. 2025 Dec 28;28:41. doi: 10.1186/s13058-025-02207-2

Subtype-specific health and economic impact of delayed breast cancer diagnosis during the early COVID-19 pandemic in Belgium: a Markov model analysis

Yasmine Khan ^1,^2,^3,^✉, Nick Verhaeghe ¹, Chris Monten ¹, Katrien Vanthomme ¹, Sylvie Gadeyne ³, Brecht Devleesschauwer ^2,⁴, Freija Verdoodt ⁵, Hanna M Peacock ⁵, Delphine De Smedt ¹

PMCID: PMC12888680 PMID: 41457254

Abstract

Background

During the first COVID-19 wave, breast cancer diagnoses declined sharply worldwide due to suspended screening programmes and delayed care-seeking driven by fear of infection. In Belgium, organized screening programmes were halted from March to June 2020, leaving 135 invasive breast cancers undiagnosed. Although no stage shifts were observed in 2020, these undiagnosed cases risk later detection at more advanced stages, with worse prognosis, higher healthcare costs, and reduced health-related quality of life. Evidence indicates that such delays disproportionately affect aggressive subtypes (e.g., triple-negative (TNBC)) compared with slower-growing luminal-like cancers. This study projected the five-year impact of diagnostic delays on health outcomes and costs, stratified by molecular subtype.

Methods

A Markov cohort model compared two cohorts of 10,147 Belgian women with breast cancer: a “disrupted-care” cohort (2020 data, including 135 undiagnosed cases) and a “non-disrupted” cohort (2017–2019 trends). Outcomes over five years were estimated from the healthcare payer perspective, including incremental QALYs, direct medical costs, and mortality. Data sources included the Belgian Cancer Registry, literature, and national cost databases. Sensitivity and scenario analyses assessed uncertainty.

Results

Over five years, diagnostic delays were projected to cause a deterministic total loss of 21 QALYs and €3.2 M in additional healthcare costs across all subtypes, resulting in six additional deaths. This corresponded to an average loss of 0.002 QALYs and €315 additional costs per patient. The burden was disproportionately carried by aggressive subtypes. TNBC accounted for the largest health loss (− 9.5 QALYs) and highest incremental costs (€1.6 M), followed by HER2+ cancer (− 2.5 QALYs; €0.5 M). Probabilistic sensitivity analysis revealed considerable uncertainty in these estimates, particularly influenced by assumed input parameters.

Conclusion

The impact of diagnostic delays during Belgium’s first COVID-19 wave was less severe than expected, likely reflecting rapid recovery measures that limited sustained stage shifts. While overall effects were modest, they may conceal a higher burden among faster-progressing subtypes such as TNBC and HER2+. Substantial model uncertainty highlights the need for routine registration of molecular subtypes to support future research on more tailored screening and diagnostic approaches during health system disruptions.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13058-025-02207-2.

Keywords: Breast cancer, COVID-19, Diagnostic delay, Molecular subtypes, Markov model, Health economics, QALY, Belgium

Background

Cancer remains a leading cause of morbidity and mortality worldwide, placing a substantial health and economic burden on societies. In Europe alone, total cancer-related costs were estimated at €199 billion in 2018 [1–3], including direct medical expenses, indirect costs (e.g., productivity losses), and direct non-medical costs like informal caregiving. Breast cancer is the most prevalent cancer among women and the second costliest cancer globally [4, 5]. Belgium has among the highest age-standardized breast cancer incidence rates in the world [6], contributing substantially to the national healthcare burden. In 2018, breast cancer healthcare costs exceeded €300 million [4]. This is driven by high incidence—11,636 new cases in 2023 [7]—and an aging population, which increases disease risk and care complexity [4]. Despite the high costs, breast cancer has a relatively low burden of years lived with disability per case [4]. This is largely due to early detection—often through organized screening programmes—and effective treatment, which improve prognosis and reduce recurrence and long-term disability [4]. These outcomes are further supported by Belgium’s structured breast cancer care pathway, with accredited clinics, mandatory multidisciplinary consultations, and national standards for timely, coordinated, high-quality care [7].

During the first wave of COVID-19, many countries implemented strict public health measures, including lockdowns, social distancing, and healthcare resource reallocation, to contain the spread of the virus [8]. These led to suspension of non-urgent services and disrupted breast cancer care globally, with halted screening and adapted treatments to reduce exposure and preserve capacity [9]. There was also concern that patients delayed seeking care due to fear of infection, healthcare strain, or a desire not to burden the system [10, 11]. Delayed breast cancer care could theoretically lead to more advanced-stage presentations, resulting in higher treatment costs, reduced health-related quality of life (HRQoL), and poorer survival [12, 13].

In Belgium, the organized breast cancer screening programme was suspended from March to June 2020, causing a 44% drop in Ductal Carcinoma In Situ (DCIS) diagnoses [14]. As DCIS is mainly detected through screening, this decline reflects programme disruption, though some were still found symptomatically. By year-end, invasive breast cancer diagnoses remained 5% below expected levels, despite screening resumption [14]. Importantly, Belgian data show that time to treatment was largely maintained or even accelerated, with no increase in tumour size or nodal involvement among symptomatic cases [14].

Breast cancer is a heterogenous disease with multiple subtypes that differ in biological behaviour and clinical outcomes. The most widely used classification system relies on immunohistochemical profiling of estrogen, progesterone, and human epidermal growth factor receptors 2 (HER2). Based on these markers, breast cancers are commonly grouped into four subtypes (Table 1): luminal A, luminal B, HER2-positive (HER2+), and triple-negative (TNBC). This classification guides treatment decisions and reflects differences in prognosis and response to therapy [15, 16]. Luminal A cancers tend to grow slowly and have a favourable prognosis, whereas HER2+ and TNBC subtypes are more aggressive, with faster progression and worse outcomes if diagnosis is delayed [11, 17]. In Belgium, data from 2014 show that luminal A-like tumours accounted for 54% of cases, followed by luminal B-like (27%), TNBC (13%), and HER2+ (6%) breast cancers [18].

Table 1.

Characteristics of the four breast cancer molecular subtypes [17]

Molecular subtype	Receptor status	Biological behaviour	Prognosis	Treatment
Luminal A	ER+, PR+, HER2–, low Ki-67	Slow-growing, hormone-sensitive	Favourable	Endocrine therapy
Luminal B	ER+, PR ± , HER2– or HER2+, high Ki-67	More proliferative than Luminal A	Intermediate	Endocrine therapy + CDK4/6 inhibitors, ± chemotherapy
HER2+	HER2+, ER–/low, PR–/low	Highly proliferative, treatment-sensitive	Variable (improved with HER2-targeted therapy)	Anti-HER2 therapy (e.g., trastuzumab, pertuzumab) + chemotherapy
TNBC	ER–, PR–, HER2–	Highly aggressive, rapid progression	Poor	Chemotherapy ± immunotherapy

Open in a new tab

ER Estrogen receptor, PR Progesterone receptor, HER2 Human Epidermal Growth Factor Receptor 2, TNBC Triple-negative breast cancer

*Ki-67 = a marker of cellular proliferation, used to distinguish Luminal A (low Ki-67) from Luminal B (high Ki-67)

**CDK4/6 = cyclin-dependent kinases 4 and 6, targeted by specific inhibitors in HR + /HER2 − breast cancer to block cell cycle progression

The impact of a temporary disruption of cancer screening on HRQoL, healthcare costs, and mortality across breast cancer molecular subtypes in Belgium remains unknown. However, evidence from other countries indicates that even short-term screening interruptions may result in more advanced-stage diagnoses [19] and higher projected mortality [20–23], particularly among aggressive subtypes. These delays can also lead to substantial long-term HRQoL and economic burdens [23, 24].

A recent study [25] found that even short diagnostic delays can lead to more advanced cancer, especially for aggressive subtypes. For TNBC, a delay of 38 days might lead to over 10% more patients being diagnosed at a later stage. This window was 52 days for HER2+ cancers but a longer 85 days for slower-growing luminal-like cancers, highlighting the heightened sensitivity of aggressive tumours to diagnostic delays.

Although Belgian data did not indicate tumour or nodal progression among breast cancer cases with delayed diagnosis between March and June 2020, these analyses were not stratified by molecular subtype [14]. As such, potential stage progression in aggressive subtypes, such as TNBC or HER2+ cancers, may have been masked by the predominance of slower-growing tumours like luminal A.

Through a Markov cohort simulation model, we explore the potential hidden impact of these diagnostic delays on changes in Quality-Adjusted Life Years (QALYs) and healthcare costs, across molecular subtype, compared to pre‑COVID‑19 care conditions. In line with Belgian guidelines for health economic evaluations [26], these standard metrics were selected to capture the comprehensive patient health burden and the direct economic impact on the healthcare system, respectively. We projected the impact over a five-year horizon, capturing the period of highest recurrence risk for aggressive subtypes such as TNBC and HER2+ [27]. This timeframe also aligns with the most robust clinical evidence on outcomes and costs [15, 16]. In addition, a 10-year horizon was examined in scenario analysis to account for the more prolonged recurrence risk observed in luminal subtypes [27]. These findings aim to provide healthcare professionals and policymakers with evidence to safeguard continuity of diagnostic services during future system shocks, minimise backlogs once services resume, prioritise timely detection of aggressive subtypes such as TNBC and HER2+ cancers, and encourage further breast cancer research stratified by molecular subtype.

Methods

We used a Markov cohort simulation model approach, as described in our protocol  [28], to project over five years the health and healthcare cost impacts of COVID-19-attributable breast cancer diagnostic disruptions that occurred from mid-March to June 2020 in Belgium. Markov models effectively capture breast cancer’s chronic nature, stage progression, and long-term recurrence and mortality risk [29]. The analysis was stratified by four breast cancer molecular subtypes (Table 1): luminal A, luminal B, HER2+, and TNBC.

Study design

Aggregated, de-identified incidence data by clinical stage were obtained from the Belgian Cancer Registry (BCR, 2024), which compiles nationwide data from oncology programmes and pathology laboratories (age, stage, tumour and patient characteristics) [30]. Ethical Committee approval was not required by law, since the collection of the data and making them available was performed within the legal framework of BCR (legal basis: art. 138 Coordinated Law of 10 May 2015), and since this retrospective study does not fall under the Belgian Law of 7 May 2004 regarding experiments on human persons (art. 3,§2) which only requires EC approval for prospective studies.

Subtype-specific Markov cohort models were developed, simulating the movement of two closed incident cohorts of 10,147 adult female breast cancer patients over five years. These included a non-disrupted care cohort, estimated from 2017–2019 trends using a Poisson model, and a disrupted-care cohort reflecting observed 2020 cases. Predicted stage-specific values were adjusted to match the observed decline, ensuring equal cohort sizes for unbiased comparison. The comparison between projected and observed data revealed a shortfall of 218 cases in Stage I, which was partially offset by slight observed excesses in Stages II–IV (see Table 2). The core of the diagnostic disruption was represented by the Undiagnosed Cancer (UDC) cohort, defined as the net deficit of 135 cases derived by subtracting the total observed stage I–IV cases in 2020 (10,012) from the total predicted incidence (10,147).

Table 2.

Breast cancer cases by stage and subtype in non-disrupted (2017–2019) and disrupted (2020) cohorts (aggregated data from the Belgian Cancer Registry, 2024)

Stage (clinical) at diagnosis	Total	Luminal A	Luminal B	HER2-positive	Triple-negative
Non-disrupted care cohort (Pre-COVID-19 situation, 2017–2019)
Stage I	4769 (47.00%)	2575	1288	286	620
Stage II	3904 (38.47%)	2108	1054	234	508
Stage III	716 (7.06%)	387	193	43	93
Stage IV	758 (7.47%)	409	205	45	99
Total	10,147	5479	2740	609	1319
Disrupted-care cohort (COVID-19 situation, 2020)
Stage I	4551 (44.85%)	2458	1229	273	592
Stage II	3949 (38.92%)	2132	1066	237	513
Stage III	753 (7.42%)	407	203	45	98
Stage IV	759 (7.48%)	410	205	46	99
Undiagnosed Cancer	135 (1.33%)	73	36	8	18
Total	10,147	5479	2740	609	1319

Open in a new tab

Values represent aggregated patient counts and have been rounded; minor discrepancies may occur

Subtype-specific counts were estimated by applying pre-COVID Belgian subtype proportions [18] to stage counts

In the absence of subtype-specific stage incidence data, we assumed that each subtype followed the overall stage distribution of breast cancer

Age stratification was not applied, as molecular subtype is the primary driver of prognosis and treatment decisions [15, 16]. While age may influence care through comorbidities or patient preference, including it would have added complexity without improving model validity. Because subtype-specific incidence data were unavailable, pre-COVID Belgian subtype proportions were applied to both cohorts by multiplying each stage count by the corresponding subtype share to estimate per-subtype counts.

A five-year time horizon was chosen to capture the period of highest recurrence risk and cost concentration, particularly for aggressive subtypes such as TNBC and HER2+ [27], and to align with follow-up guidelines and available evidence. A three-month cycle length was used, reflecting typical treatment intervals (e.g., chemotherapy cycles, clinical follow-ups) [15]. The model adopted a Belgian healthcare payer perspective, focusing on direct medical costs (e.g., treatments, consultations). This approach aligns with the National Institute for Health and Disability Insurance (NIHDI) reimbursement practice and avoids the added uncertainty that would arise from applying a societal perspective, given the lack of subtype-specific data on productivity losses and informal caregiving.

The model was developed in Excel, with patients distributed at baseline (cycle 0) across the initial health states—‘Stage I Diagnosis &Treatment (D&T)’, ‘Stage II D&T’, ‘Stage III D&T’, ‘Stage IV D&T’, and ‘UDC’ (see Fig. 1)—according to BCR incidence data. Each cohort then progressed through health states via a transition matrix, which applied predefined probabilities to redistribute patients every three months. Each health state was associated with specific healthcare costs and utilities (see 2.4 Data inputs). By iterating these cycles, the simulation generated a dynamic representation of the patient population’s trajectory over the five-year horizon. Ultimately, the model compared QALYs (where 1 = perfect health and 0 = death) and healthcare costs between cohorts. Input parameters for the simulation, including transition probabilities, healthcare costs, and utilities, are detailed in Table 4 (presented later in the manuscript), with calculation methods provided in the Additional file 1.

Fig. 1 — Breast cancer Markov model diagram with 11 health states, applicable to each molecular subtypes. D & T: Diagnosis and treatment; FU: Follow-up; L: Lines of systemic therapy; LRR: Locoregional recurrence. L: Lines of systemic therapy; UDC: Undiagnosed cancer. ↺ = Probability of remaining in the state. → = Probability of moving to another state.] = Probability for patients in any state to move to the dead state

Table 4.

Baseline input parameters for non-disrupted care (pre-COVID-19) and disrupted care (COVID-19) cohorts by molecular subtype

Input parameters	Subtype	Base Value	Standard Error	OSA—95% CI Lower Bound	OSA—95% CI Upper Bound	Source(s)
Transition probabilities
Stage I D & T to Dead	Luminal A	0.002	0.001	0.001	0.004	[81]
	Luminal B	0.010	0.002	0.006	0.013
	HER2+	0.016	0.002	0.012	0.021
	TNBC	0.017	0.002	0.012	0.02
Stage II D & T to Dead	Luminal A	0.005	0.001	0.002	0.007
	Luminal B	0.012	0.002	0.008	0.016
	HER2+	0.019	0.003	0.014	0.024
	TNBC	0.019	0.003	0.014	0.024
Stage III D & T to Dead	Luminal A	0.012	0.002	0.008	0.016
	Luminal B	0.020	0.003	0.015	0.025
	HER2+	0.026	0.003	0.020	0.032
	TNBC	0.027	0.003	0.021	0.032
Stage I–III FU to Stage I-III LRR	Luminal A	0.003	0.001	0.001	0.004	[27]
	Luminal B	0.005	0.002	0.002	0.008
	HER2+	0.006	0.003	0.000	0.013
	TNBC	0.007	0.003	0.002	0.012
Stage I–III FU to Stage IV D & 1L	Luminal A	0.005	0.001	0.003	0.007
	Luminal B	0.011	0.002	0.007	0.016
	HER2+	0.015	0.005	0.004	0.025
	TNBC	0.013	0.004	0.006	0.020
Stage I–III FU to Dead	Luminal A	0.001	0.000	0.000	0.001
	Luminal B	0.001	0.001	0.000	0.002
	HER2+	0.001	0.002	0.000	0.006
	TNBC	0.002	0.001	0.000	0.004
Stage I–III LRR to Dead	Luminal A	0.011	0.002	0.008	0.014	[82]
	Luminal B	0.032	0.005	0.023	0.042
	HER2+	0.020	0.004	0.013	0.028
	TNBC	0.035	0.006	0.023	0.046
Stage IV D & 1L to Stage IV 2L	Luminal A	0.060	0.005	0.051	0.069	[83]
	Luminal B	0.060	0.005	0.051	0.069
	HER2+	0.054	0.008	0.039	0.069
	TNBC	0.062	0.012	0.038	0.085
Stage IV D & 1L to Dead	Luminal A	0.005	0.001	0.003	0.008
	Luminal B	0.005	0.001	0.003	0.008
	HER2+	0.007	0.004	0.000	0.014
	TNBC	0.012	0.005	0.002	0.023
Stage IV 2L to Stage IV 3L	Luminal A	0.073	0.006	0.061	0.085
	Luminal B	0.073	0.006	0.061	0.085
	HER2+	0.065	0.010	0.045	0.086
	TNBC	0.065	0.014	0.037	0.093
Stage IV 2L to Dead	Luminal A	0.008	0.002	0.004	0.012
	Luminal B	0.008	0.002	0.004	0.012
	HER2+	0.007	0.004	0.000	0.014
	TNBC	0.013	0.007	0.000	0.026
Stage IV 3L to Stage IV 4L	Luminal A	0.073	0.007	0.059	0.087
	Luminal B	0.073	0.007	0.059	0.087
	HER2+	0.063	0.013	0.039	0.088
	TNBC	0.057	0.016	0.026	0.087
Stage IV 3L to Dead	Luminal A	0.009	0.003	0.004	0.014
	Luminal B	0.009	0.003	0.004	0.014
	HER2+	0.009	0.005	0.000	0.018
	TNBC	0.018	0.009	0.000	0.035
Stage IV 4L to Dead	Luminal A	0.010	0.003	0.004	0.015
	Luminal B	0.010	0.003	0.004	0.015
	HER2+	0.010	0.010	0.000	0.031
	TNBC	0.022	0.010	0.003	0.041
UDC to Stage I D & T	Luminal A	0.400	0.020	0.361	0.439	Hypothesized based on clinical evidence of stage progression after diagnostic delay [25]
	Luminal B	0.300	0.020	0.261	0.389
	HER2+	0.100	0.020	0.061	0.139
	TNBC	0.050	0.020	0.011	0.089
UDC to Stage II D & T	Luminal A	0.400	0.020	0.361	0.439
	Luminal B	0.350	0.020	0.311	0.389
	HER2+	0.150	0.020	0.111	0.189
	TNBC	0.050	0.020	0.011	0.089
UDC to Stage III D & T	Luminal A	0.100	0.020	0.061	0.139
	Luminal B	0.150	0.020	0.111	0.189
	HER2+	0.300	0.020	0.261	0.339
	TNBC	0.100	0.020	0.061	0.139
UDC to Stage IV D & 1L	Luminal A	0.080	0.020	0.041	0.119
	Luminal B	0.150	0.020	0.111	0.189
	HER2+	0.350	0.020	0.311	0.389
	TNBC	0.750	0.020	0.711	0.789
UDC to Dead	Luminal A	0.010	0.020	0.000	0.049
	Luminal B	0.025	0.020	0.000	0.064
	HER2+	0.035	0.020	0.000	0.074
	TNBC	0.040	0.020	0.001	0.079

Input parameters	Subtype	Base Value	Standard Error	Disutility value	OSA—95% CI Lower Bound	OSA—95% CI Upper Bound	Source(s)
Utilities
UDC	Luminal A	0.82	0.005	-	0.811	0.829	[36]
	Luminal B	0.82	0.005		0.811	0.829
	HER2+	0.82	0.005		0.811	0.829
	TNBC	0.82	0.005		0.811	0.829
Stage I D & T	Luminal A	0.787	0.022	0.033	0.744	0.829	[37]
	Luminal B	0.787	0.022	0.033	0.744	0.829
	HER2+	0.770	0.077	0.050	0.620	0.920	[41–43]
	TNBC	0.755	0.077	0.065	0.605	0.905	[43, 44]
Stage II D & T	Luminal A	0.787	0.022	0.033	0.744	0.829	[37]
	Luminal B	0.787	0.022	0.033	0.744	0.829
	HER2+	0.770	0.077	0.050	0.620	0.920	[41–43]
	TNBC	0.755	0.077	0.065	0.605	0.905	[43, 44]
Stage III D &T	Luminal A	0.782	0.025	0.038	0.733	0.831	[37]
	Luminal B	0.782	0.025	0.038	0.733	0.831
	HER2+	0.760	0.077	0.060	0.610	0.910	[41–43]
	TNBC	0.745	0.077	0.075	0.95	0.895	[43, 44]
Stage I–III FU (adjuvant therapy)	Luminal A	0.772	0.021	0.048	0.731	0.812	[37]
	Luminal B	0.772	0.021	0.048	0.731	0.812
	HER2+	0.733	0.041	0.087	0.654	0.813
	TNBC	0.733	0.041	0.087	0.654	0.813
Stage I–III FU (post-adjuvant therapy)	Luminal A	0.814	0.005	0.006	0.805	0.823	[37]
	Luminal B	0.814	0.005	0.006	0.805	0.823
	HER2+	0.814	0.005	0.006	0.805	0.823
	TNBC	0.814	0.005	0.006	0.805	0.823
Stage I–III LRR	Luminal A	0.770	0.007	0.050	0.757	0.783	[38]
	Luminal B	0.770	0.007	0.050	0.757	0.783
	HER2+	0.730	0.077	0.090	0.580	0.880
	TNBC	0.710	0.077	0.110	0.560	0.860
Stage IV D & T	Luminal A	0.730	0.009	0.090	0.712	0.748	[38]
	Luminal B	0.730	0.009	0.090	0.712	0.748
	HER2+	0.690	0.077	0.130	0.540	0.840	[41]
	TNBC	0.670	0.077	0.150	0.520	0.820	[47]
Stage IV 2L	Luminal A	0.720	0.077	0.100	0.570	0.870	[48]
	Luminal B	0.720	0.016	0.100	0.570	0.870
	HER2+	0.680	0.077	0.140	0.530	0.830	[49]
	TNBC	0.650	0.077	0.170	0.500	0.800	[50]
Stage IV 3L	Luminal A	0.700	0.077	0.120	0.550	0.850	[49]
	Luminal B	0.700	0.077	0.120	0.550	0.850
	HER2+	0.680	0.077	0.140	0.530	0.830	[51]
	TNBC	0.630	0.077	0.190	0.480	0.780	[52]
Stage IV 4L	Luminal A	0.690	0.077	0.130	0.540	0.840	[53, 84]
	Luminal B	0.690	0.077	0.130	0.540	0.840
	HER2+	0.660	0.077	0.160	0.510	0.810	[54]
	TNBC	0.600	0.077	0.220	0.450	0.750	[55]

Input parameters	Subtype	Base Value	Standard Error	OSA—95% CI Lower Bound	OSA—95% CI Upper Bound	Source(s)
Direct healthcare costs
Stage I D & T	Luminal A	4,548	455	3,657	5,440	NomenSoft nomenclature codes (see the Additional file 1)
	Luminal B	4,548	455	3,657	5,440
	HER2+	7,563	756	6,081	9,046
	TNBC	28,352	2,835	22,795	33,910
Stage II D & T	Luminal A	4,548	455	3,657	5,440
	Luminal B	4,548	455	3,657	5,440
	HER2+	12,601	1,260	10,131	15,071
	TNBC	28,352	2,835	22,795	33,910
Stage III D &T	Luminal A	5,523	552	4,441	6,606
	Luminal B	5,523	552	4,441	6,606
	HER2+	13,576	1,358	10,915	16,236
	TNBC	29,327	2,933	23,579	35,075
Stage I–III FU (adjuvant therapy)	Luminal A	536	54	431	641	BCPI [34]
	Luminal B	536	54	431	641
	HER2+	16,105	1,610	12,948	19,262
	TNBC	23,804	2,380	19,139	28,470
Stage I–III FU (post-adjuvant therapy)	Luminal A	334	33	269	400	NomenSoft nomenclature codes (see the Additional file 1)
	Luminal B	334	33	269	400
	HER2+	334	33	269	400
	TNBC	334	33	269	400
Stage I–III LRR	Luminal A	4,873	487	3,918	5,828	Mean Stage I-III D&T costs
	Luminal B	4,873	487	3,918	5,828
	HER2+	11,247	1,125	9,042	13,451
	TNBC	28,677	2,868	23,057	34,298
Stage IV D & T	Luminal A	5,934	593	4,771	7,098	BCPI [34]
	Luminal B	5,934	593	4,771	7,098
	HER2+	16,385	1,638	13,173	19,596
	TNBC	24,649	2,465	19,818	29,480
Stage IV 2L	Luminal A	5,922	592	4,762	7,083
	Luminal B	5,922	592	4,762	7,083
	HER2+	14,537	1,454	11,688	17,386
	TNBC	25,038	2,504	20,131	29,945
Stage IV 3L	Luminal A	28,976	2,898	23,297	34,655
	Luminal B	28,976	2,898	23,297	34,655
	HER2+	4,768	477	3,833	5,702
	TNBC	9,374	937	7,537	11,211
Stage IV 4L	Luminal A	1,261	126	1,014	1,509
	Luminal B	1,261	126	1,014	1,509
	HER2+	19,663	1,966	15,809	23,518
	TNBC	1,245	124	1,001	1,488

Input parameters	Subtype	Base Value	Standard Error	Scenario Analysis—Multiplier	OSA—95% CI Lower Bound	OSA—95% CI Upper Bound	Source(s)
Scenario analysis
UDC to Stage I D & T	Luminal A	0.10	0.001	0.25	0.098	0.102	Hypothesized based on clinical evidence of stage progression after diagnostic delay [25]
	Luminal B	0.09	0.001	0.30	0.088	0.092
	HER2+	0.05	0.001	0.50	0.049	0.051
	TNBC	0.01	0.000	0.20	0.009	0.011
UDC to Stage II D & T	Luminal A	0.10	0.001	0.25	0.098	0.102
	Luminal B	0.12	0.001	0.35	0.120	0.125
	HER2+	0.05	0.001	0.35	0.051	0.054
	TNBC	0.01	0.000	0.20	0.009	0.011
UDC to Stage III D & T	Luminal A	0.25	0.001	2.50	0.247	0.253
	Luminal B	0.30	0.001	2.00	0.297	0.303
	HER2+	0.35	0.002	1.15	0.342	0.348
	TNBC	0.05	0.001	0.50	0.039	0.041
UDC to Stage IV D & 1L	Luminal A	0.52	0.002	6.50	0.517	0.523
	Luminal B	0.45	0.002	3.00	0.447	0.453
	HER2+	0.46	0.002	1.30	0.452	0.458
	TNBC	0.86	0.001	1.15	0.860	0.865
UDC to Dead	Luminal A	0.01	0.000	1.00	0.009	0.011
	Luminal B	0.03	0.000	1.00	0.024	0.026
	HER2+	0.04	0.001	1.00	0.034	0.036
	TNBC	0.04	0.001	1.00	0.039	0.041

Open in a new tab

D: Diagnosis; FU: Follow-Up; LRR: Locoregional Recurrence; OSA: One-way Sensitivity Analysis; T: Treatment; TNBC: Triple-Negative Breast Cancer; UDC: Undiagnosed Cancer; 1L: 1st systemic therapy line; 2L: 2nd systemic therapy line; 3L: 3rd systemic therapy line; 4L: 4th systemic therapy line

*Bold input parameters represent those only present during the disrupted care situation (COVID-19)

**95% CI: 95% confidence interval, which was calculated as follows: Base case value ± 1.96 × SE

***Standard errors were calculated using Inline graphic for means and for proportions

****Hypothesized input values were underlined

Markov model structure

The model consisted of 11 health states (Fig. 1), grouped into non-metastatic (stages I–III) and metastatic (stage IV) disease. Non-metastatic disease begins with a 'diagnosis and treatment' (D&T) phase, which includes diagnostic work-up, neoadjuvant therapy (for HER2+ and TNBC), surgery (with reconstruction if applicable), and radiotherapy. This phase is represented in the model as 'Stage I D&T,' 'Stage II D&T,' or 'Stage III D&T,' depending on the initial stage at diagnosis. Patients remain in the D&T state for one or more cycles depending on their subtype-specific treatment duration (Table 3), before transitioning to 'Stage I–III Follow-Up (FU)' for a defined period of adjuvant therapy (AT). We modelled AT as five years (20 cycles) for luminal A and seven years (28 cycles) for luminal B, reflecting standard practice for hormone receptor–positive disease and extended therapy increasingly recommended for luminal B due to its higher risk of late recurrence [31].

Table 3.

Duration spent in initial treatment-related health states by molecular subtype, independent of COVID-19 disruptions

Health States	Luminal A	Luminal B	HER2 +	Triple-Negative
Stage I D & T	3 months (1 cycle)	3 months (1 cycle)	9 months (3 cycles)	9 months (3 cycles)
Stage II D & T	3 months (1 cycle)	3 months (1 cycle)	9 months (3 cycles)	9 months (3 cycles)
Stage III D & T	3 months (1 cycle)	3 months (1 cycle)	9 months (3 cycles)	9 months (3 cycles)
Stage I-III FU (Adjuvant therapy)	5 years (20 cycles)	7 years (28 cycles)	1 year (4 cycles)	1 year (4 cycles)

Open in a new tab

D&T diagnosis and treatment, FU follow-up

Stage IV (metastatic) disease is not included in this table, as its clinical course typically involves successive lines of systemic therapy and differs from the more structured, stage-based pathways of earlier stages. While treatment-free or even disease-free intervals may occur, these patterns were not incorporated in the model for reasons of simplification. The durations listed here reflect initial, curative-intent treatment for non-metastatic disease (stages I–III) prior to progression

Given the differing nature and duration of AT across molecular subtypes, a distinct ‘Stage I–III FU’ health state was defined. After AT, patients enter this state with follow-up every three months until recurrence. Some, however, may not achieve a disease-free period and directly experience disease progression. In both cases, we simplify this by saying that patients move to ‘Stage I–III Locoregional Recurrence (LRR)’ if the evolution is locoregional or to ‘Stage IV D&T’ if metastatic. Those in ‘Stage I–III LRR’ receive treatment during one cycle and then return to follow-up, while patients entering the metastatic pathway (Stage IV states) start systemic therapy in ‘Stage IV D&T’ and transition through subsequent lines of treatment as disease advances.

Patients diagnosed de novo with metastatic disease also start at ‘Stage IV D&T’, where they receive a first line of systemic therapy. They remain in this state (illustrated through the looping arrow) until disease progression, where they directly transition to the second-line regimen. This process continues sequentially up to four lines of metastatic treatment: ‘Stage IV 2L’, ‘3L’, and ‘4L’. A separate health state, ‘Undiagnosed Cancer’ (UDC), captures patients whose diagnosis was delayed due to COVID-19-related disruptions and who can subsequently transition to any stage at diagnosis once detected. During any cycle, patients in any health state can transition to ‘Dead’ (absorbing state).

Model validation

Clinical validation was conducted through consultation rounds with oncologists C.M. and E.N., to validate the model structure (i.e., model diagram), input parameters (i.e., healthcare costs), and key assumptions (i.e., simplified treatment pathways), ensuring alignment with current clinical practice and expert opinion. Additionally, five‑ and ten‑year mortality outcomes from the non-disrupted care model were validated against published subtype‑specific survival rates [32], converted to mortality values (mortality = 1 – survival), which were consistent with our estimates.

Data inputs

To operationalise our Markov model, aggregated incidence data from the BCR were used, transition probabilities and utilities were extracted from the literature, and healthcare costs were sourced from the Belgian medical nomenclature database (NomenSoft) and the Belgian pharmacotherapeutic information database (BCPI). All input values were stratified by health state and molecular subtype. The selected input values are presented in Table 4, with calculation methods and rationale detailed in the Additional file 1.

Both models used the same input values, except for aggregated incidence data and transition probabilities from the UDC state to stage I–IV D&T. In the non-disrupted care cohort, there were no undiagnosed breast cancers related to care disruptions, and thus no transitions from the UDC state occurred. In contrast, the disrupted care model included UDC cases with hypothesized transition probabilities to stage I–IV D&T to reflect diagnostic delays.

Epidemiological data

As mentioned above, stage-specific incidence data for breast cancer diagnoses for the disrupted and non-disrupted care cohorts were obtained from the BCR, enabling the estimation of UDC (they represent stage-specific differences between cohorts). In the absence of subtype-specific stage incidence data, we assumed that each subtype followed the overall stage distribution of breast cancer, applying pre-COVID Belgian subtype proportions [18] to derive per-subtype estimates. Under 5% of diagnoses in the BCR dataset lacked definitive staging (unknown stage). Reasons for incomplete staging are hypothesized to differ between non-disrupted care cohort (e.g., frail elderly) and disrupted care cohort (e.g., potentially, interrupted diagnostics due to intensive care unit admission). Given the uncertainty in distribution, the minimal overall percentage, and the absence of a statistically significant difference between cohorts, unknown stage cases were excluded from the main model to simplify comparisons.

Transition probabilities

The 3-month subtype-specific transition probabilities were mainly derived from international retrospective cohort studies (Table 4). We hypothesized 3-month transition probabilities from UDC to ‘Stage I–IV D&T’ and ‘Dead’ for each molecular subtypes—given the absence of subtype-specific incidence data on UDC—based on a study which showed that aggressive subtypes (i.e., TNBC and HER2+) reach stage progression much earlier than luminal-like cancers [25].

Healthcare costs

Healthcare costs for each health state were estimated from the perspective of the Belgian healthcare payer (NIHDI). The model's costs represent the official reimbursed amounts (nomenclature codes 1300/1600). Patient co-payments (codes 3300/3600) were excluded to reflect only the payer's expenditure. Diagnostic, surgical, reconstruction, and radiotherapy costs were retrieved from NomenSoft [33], which provides official reimbursement rates (codes are available in the Additional file 1). Subtype-specific drug costs for neoadjuvant and AT were obtained from the Belgian Centre for Pharmacotherapeutic Information (BCPI) [34], with dosages and schedules based on phase III trials and validated by two oncologists (C.M., E.N.). All reimbursement values were indexed to 2020 prices using the Belgian health index [35] to ensure consistency with the model’s reference year.

Subtype-specific treatment pathways (Table 5) were simplified to balance between typical clinical practice and ensure feasibility within the Markov framework, and were based on ESMO clinical practice guidelines [15, 16], which underpin Belgian national recommendations. Some treatment regimens were generalized, and treatment durations were averaged across patient groups. A standardized diagnostic pathway was assumed across all molecular subtypes, including biopsy, staging imaging, blood analysis, and mammography. For simplicity, we assumed only patients with HER2+ and TNBC to have subtype-specific neoadjuvant therapy. Surgery was stratified by stage: All stage I-II patients underwent breast-conserving surgery (BCS) without reconstruction, while for stage III patients, 70% underwent BCS and 30% underwent mastectomy, with 25% assumed to receive reconstruction. For simplicity, surgery was assumed to remain unchanged during the pandemic. All patients received the standard hypofractionated whole-breast radiotherapy regimen of 15 fractions, and related consultations before, during, and after treatment. AT also differed based on the molecular subtype. Post-adjuvant therapy follow-up included annual mammography, ultrasonography, and four consultations per year. For stage I-III patients developing LRR, healthcare costs were assumed to mirror those of initial D&T, calculated as the mean cost across stages I-III. Metastatic disease treatments followed subtype-specific systemic therapy lines, with healthcare costs also estimated on a 3-month cycle basis for comparability across subtypes. For simplicity, luminal B was assumed to have the same costs as luminal A.

Table 5.

Simplified treatment pathway across health states and molecular subtype

Health state	Luminal-like	HER2 +	TNBC
Stage I D & T	Diagnosis + BCS + RT (+ pre, per, post consultations)	Diagnosis + Neoadjuvant therapy + BCS + RT (+ pre, per, post consultations)	Diagnosis + Neoadjuvant therapy + BCS + RT (+ pre, per, post consultations)
Stage II D & T	Diagnosis + BCS + RT (+ pre, per, post consultations)	Diagnosis + Neoadjuvant therapy + BCS + RT (+ pre, per, post consultations)	Diagnosis + Neoadjuvant therapy + BCS + RT (+ pre, per, post consultations)
Stage III D & T	Diagnosis + 70% BCS and 30% MST (+ 25% reconstruction) + RT (including pre, per, post consultations)	Diagnosis + Neoadjuvant therapy + 70% BCS and 30% MST (+ 25% reconstruction) + RT (including pre, per, post consultations)	Diagnosis + Neoadjuvant therapy + 70% BCS and 30% MST (+ 25% reconstruction) + RT (including pre, per, post consultations)
Stage I-III FU (Adjuvant therapy)	Adjuvant endocrine therapy: Either: tamoxifen/letrozole/ Exemestane/ anastrozole + triptorelin/goserelin (if premenopausal women) + CDK4 inhibitors: Abemaciclib	Chemotherapy + Anti-HER2 therapy (+ Endocrine therapy if HER2 +)	Immunotherapy and Chemotherapy
Stage I-III FU (Post-adjuvant therapy)	Mammography Ultrasonography 4 consultations	Mammography Ultrasonography 4 consultations	Mammography Ultrasonography 4 consultations
Stage I-III LRR	Same as ‘Stage I-III D&T’	Same as ‘Stage I-III D&T’	Same as ‘Stage I-III D&T’
Stage IV D & T	Diagnosis + endocrine therapy and CDK4/6 inhibitors	Anti-HER2 and Chemotherapy	Immunotherapy and Chemotherapy
Stage IV 2L	Targeted therapy	Anti-HER2	Targeted therapy
Stage IV 3L	Chemotherapy for patients with endocrine resistance	Chemotherapy and Anti-HER2	Chemotherapy
Stage IV 4L	Chemotherapy for patients with endocrine resistance	Chemotherapy and Anti-HER2	Chemotherapy

Open in a new tab

BCS breast-conserving surgery, CDK4/6 inhibitors cyclin-dependent kinases 4 and 6, D&T Diagnosis and treatment, FU Follow-up, HER2 Human epidermal growth factor receptor 2–positive, LRR Locoregional recurrence, MST Mastectomy, RT Radiotherapy

The breast cancer Markov model Excel file (Supplementary material), includes a detailed sheet with the exact treatment cost calculations

Utilities

Health-state utilities were derived from EQ-5D-5L data [36–38] using a two-step approach. Baseline utilities for Belgian women of all ages were taken from the national Health Interview Survey [36]. Health-state-specific disutilities from international studies  [37, 38] were then subtracted, assuming comparable disease impacts across similar countries (e.g., the Netherlands). When no published disutilities were available for specific health states or subtypes, values were estimated based on additional literature [39–55], considering subtype-specific treatment burden, disease progression, and HRQoL changes. Utilities for luminal B were assumed to be equal to those of luminal A. Utilities for the UDC reflected general population levels, as asymptomatic patients were undiagnosed and likely unaffected by disease-specific HRQoL.

Analysis

By running both cohorts, incremental differences in QALYs and healthcare costs between the disrupted care and non-disrupted care situations were estimated across subtypes over a five-year time horizon. QALYs were estimated by multiplying the utility value of each health state by the number of patients in that state per cycle and the cycle length. Summing these values across all cycles yielded the total QALYs. Healthcare costs were calculated similarly, by applying cost estimates to each state and aggregating them over all cycles.

To reflect present-day valuations, both healthcare costs and QALYs were discounted annually according to Belgian guidelines—3.0% for costs and 1.5% for health effects [26]. Additionally, the analyses were used to calculate the percentage point differences in mortality and the estimated number of additional deaths attributable to delayed diagnosis across molecular subtypes.

Scenario and sensitivity analyses

A one-way sensitivity analysis (OSA) and a probabilistic sensitivity analysis (PSA) were performed to assess parameter uncertainty and evaluate the robustness of results across each molecular subtype. The OSA included transition probabilities, utilities, and direct medical costs. Each parameter was varied individually across a range defined by its 95% confidence interval, calculated as the base-case value ± 1.96 × standard error (SE), while holding all other parameters constant. Two tornado diagrams were produced: one showing the impact of varying transition probabilities and utilities on incremental QALYs, and another showing the impact of varying transition probabilities and treatment costs on incremental healthcare costs. Both diagrams reflect results combined across all molecular subtypes.

For the PSA, beta distributions were applied to probabilities and utilities, and gamma distributions to costs. Distribution parameters were derived from base-case values and their SEs, calculated from reported sample sizes where available. For parameters without reported variance, plausible 95% ranges (for probabilities/utilities) or an assumed coefficient of variation of 0.10 (for costs) were used to reflect uncertainty. The PSA was conducted using Monte Carlo simulation with 1,000 iterations in Visual Basic for Applications (VBA) within Excel.

Three scenario analyses were conducted to assess structural and methodological assumptions. A 10-year time horizon was used to explore longer-term outcomes. This captures the early recurrence peaks in HER2+ and TNBC (around two years) as well as the slower recurrence pattern of luminal A and B (beyond 10 years) [27, 56, 57]. The same utility and cost inputs per cycle were maintained, with adapted transition probabilities reflecting recurrence risks beyond five years [27]. In the second scenario, we adjusted the base-case 3-month transition probabilities from the UDC state to “Stage I–IV D&T” for each molecular subtype. The base case assumed slower-growing subtypes (e.g., luminal A) were more often diagnosed at early stages, and aggressive subtypes (e.g., TNBC) at later stages, based on clinical evidence [12, 25]. This scenario applied subtype-specific multipliers (Table 4) to model more severe progression, increasing late-stage and reducing early-stage diagnoses. As HER2+ and TNBC already had high late-stage probabilities, adjustments were minor, whereas larger multipliers were applied to luminal-like subtypes to test possible underestimation of progression severity. The third scenario examined the impact of changes in radiotherapy reimbursement tariffs during the COVID-19 pandemic. As part of emergency response strategies, several Belgian hospitals adopted hypofractionated radiotherapy (five fractions instead of 15) to reduce hospital visits, minimise infection risks, and optimise resource use [58]. This protocol was reimbursed under a higher-cost category (€2,581.73—nomenclature code: 444,710–444721) rather than the standard category III (€1,999.67 in 2020—nomenclature code: 444,150–444161). Belgian data indicate that 8.2% of patients received five fractions in 2020 [14]. Exploring this scenario is relevant to capture the potential cost implications of shifts in clinical practice and reimbursement during the pandemic.

Results

Base case analysis

The Markov model estimated that COVID-19 related diagnostic delays in breast cancer across Belgium from March-June 2020, led to a total loss of 20.7 QALYs and €3.19 million in additional healthcare costs over five years, compared to a situation without disruption (Table 6). This corresponds to an average loss of €315 per patient and 0.002 QALYs per patient—representing less than one day of life in full health (far below the minimal clinically important difference (MCID) typically cited for oncology, around 0.03 QALYs per year per patient [59]). Deterministic model outputs are point estimates derived from fixed parameter values and do not yield confidence intervals; uncertainty in these estimates was instead explored through the PSA.

Table 6.

Base case: Five-year QALY and cost changes by breast cancer subtype under care disruption

Molecular subtype	Cohort (n)	Non-disrupted care (2017–2019)			Disrupted-care (2020)			Incremental costs & QALYs (Disrupted-care vs. Non-disrupted care)
Molecular subtype	Cohort (n)	QALYs	Cost (€)	Death %	QALYs	Cost (€)	Death %	∆QALYs	∆Cost (€)	∆% Point	Additional deaths
Luminal A	Total cohort (n = 5479)	20,987.5	179,522,108	2.80	20,982.7	180,004,052	2.88	− 4.8	481,944	0.08	2
	Per patient	3.831	32,765		3.830	32,853		− 0.001	88
Luminal B	Total cohort (n = 2740)	10,378.9	103,245,734	4.04	10,375.0	103,770,423	4.09	− 4.0	524, 689	0.05	1
	Per patient	3.788	37,681		3.786	37,872		− 0.001	191
HER2 +	Total cohort (n = 609)	2, 242.4	82, 155, 748	8.06	2, 240.0	82, 702, 749	8.13	− 2.5	547, 000	0.07	1
	P,r patient	3.682	134,903		3.678	135,801		− 0.004	898
TNBC	Total cohort(n = 1319)	4,764.4	319,433,158	10.81	4,754.9	321,072,481	11.01	− 9.5	1,639,323	0.20	2
	Per patient	3.612	242,178		3.605	243,421		− 0.007	1,243
Breast cancer	Total cohort(n = 10,147)		−					-20.7	3,192,956
	Per patient		−					-0.002	315

Open in a new tab

HER2 + Human Epidermal Growth Factor Receptor 2–positive, QALY Quality-Adjusted Life Year, TNBC Triple-Negative Breast Cancer

***bold values indicate incremental outcomes comparing disrupted care (2020) with non-disrupted care (2017–2019), representing the health and economic impact attributable to COVID-19–related breast cancer diagnosis disruption

Among molecular subtypes, TNBC accounted for the largest QALY loss (− 9.5), followed by luminal A (− 4.8), luminal B (− 3.9), and HER2+ (− 2.5). TNBC patients also incurred the highest additional direct healthcare costs (€1.64 million), followed by HER2+ (€547,000), luminal B (€524,689), and luminal A (€481,944). Per patient, the largest QALY loss was observed in TNBC (− 0.007 QALYs, or 2.56 days of full health), followed by HER2+ (− 0.004 QALYs, or 1.46 days of full health), and the luminal-like subtypes, both below one day of full health (− 0.001 QALYs). TNBC patients incurred the highest additional per-patient costs (€1243), followed by HER2+ (€898), luminal B (€191), and luminal A (€88).

Delayed cancer diagnoses resulted in a slight increase in mortality across all subtypes. The largest absolute increase in mortality occurred in TNBC (+0.20 percentage points), resulting in two additional deaths. Luminal A and B subtypes showed increases in mortality of 0.08 and 0.05 percentage points, corresponding to two and one additional death(s), respectively. HER2+ increased by 0.07 percentage points, also leading to one additional death. For TNBC, this corresponds to a 1.9% relative increase compared to non-disrupted care. Luminal A and B showed relative increases of 2.9% and 1.2%, respectively, while HER2+ increased by 0.9%. Relative changes in mortality were calculated as the percentage difference from the non-disrupted care situation.

One-way sensitivity analysis (OSA)

This section presents the OSA results examining how variations in transition probabilities, costs, and utilities influence both changes in incremental QALYs and healthcare costs. Two tornado diagrams (Figs. 2 and 3) are provided to illustrate the parameters driving the greatest changes.

Fig. 2 — Tornado diagram showing the impact of varying transition probabilities and utilities on incremental QALYs across breast cancer molecular subtypes. L: Lines of systemic therapy (2L: second therapy line; 3L: third therapy line; 4L: fourth therapy line); LRR: Locoregional recurrence; LUM A: Luminal A; LUMA B: Luminal B; QALY: Quality-Adjusted Life Year; TNBC: Triple-negative breast cancer; TP: Transition probabilities; UDC: Undiagnosed cancer. * The vertical dashed line represents no change from the base case (ΔQALY = 0). Dark grey bars indicate the change when the parameter is set to its lower bound, and light grey bars indicate the change when set to its upper bound.

Fig. 3 — Tornado diagram showing the impact of varying transition probabilities and treatment costs on incremental healthcare costs across breast cancer molecular subtypes. D&T: Diagnosis and treatment; L: Lines of systemic therapy (2L: second therapy line; 3L: third therapy line; 4L: fourth therapy line); LRR: Locoregional recurrence; LUM A: Luminal A; LUMA B: Luminal B; TNBC: Triple-negative breast cancer; TP: Transition probabilities; UDC: Undiagnosed cancer. *The vertical dashed line represents no change from the base case (Δ healthcare costs = 0). Dark grey bars indicate the change when the parameter is set to its lower bound, and light grey bars indicate the change when set to its upper bound

The impact of varying transition probabilities and utilities on incremental QALYs

The OSA on QALYs showed that the model was most sensitive to changes in the utilities for TNBC. Particularly, utilities for stage I D&T, stage I–III adjuvant therapy, stage IV D&T, and stage II D&T had the largest impact on incremental QALYs.

The impact of varying model inputs on incremental healthcare cost changes

The OSA on healthcare costs showed that the model was most sensitive to cost variations in TNBC. Specifically, the costs of stage I D&T and stage IV D&T, which had the largest impacts on total incremental healthcare costs.

Scenario analyses

Extending the time horizon from five to 10 years revealed a more pronounced impact of COVID-19-related disruptions on both health and economic outcomes (Table 7). QALY losses widened from − 20.7 to − 32.1, and incremental healthcare costs rose by €2.50 million (from €3.19 million to €5.69 million). Luminal-like subtypes experienced over a twofold increase in both QALY losses and healthcare costs, whereas increases were less pronounced for TNBC and HER2+ : QALY losses doubled for HER2+ but fell by more than half for TNBC, while costs rose by over 1.5-fold for both. Mortality trends were consistent with the base case, with TNBC showing the largest absolute increase (+ 0.33 percentage points), followed by HER2+ (+ 0.11), luminal B (+ 0.06), and luminal A (+ 0.04). These corresponded to relative increases in mortality of 1.6% for TNBC (from 20.44 to 20.77%), 0.8% for HER2+ (from 13.64 to 13.75%), 0.7% for luminal B (from 9.10 to 9.16%), and 0.6% for luminal A (from 6.36 to 6.40%).

Table 7.

Scenario analysis: Incremental changes in QALYs and healthcare costs across three scenarios, by molecular subtype

Scenario type	Molecular Subtype	Cohort (n)	Non-disrupted care (2017–2019)			Disrupted care (2020)			Incremental healthcare costs & QALYs (Disrupted care vs. Non-disrupted care)
Scenario type	Molecular Subtype	Cohort (n)	QALYs	Cost (€)	Death %	QALYs	Cost (€)	Death%	∆QALYs	∆Cost (€)	∆% Point	Additional deaths
10-year time horizon	Luminal A	Total	39,531.7	342,044,246	6.36	39,517.6	343,230,607	6.40	− 13.1	1,186,360	0.04	3
		Per patient	7.215	62,428		7.213	62,645		− 0.002	217
	Luminal B	Total	19,182.2	228,963,281	9.10	19,172.2	230,020,328	9.16	− 10.0	1,057,047	0.06	2
		Per patient	7.001	83,563		6.997	83,949		− 0.004	386
	HER2+	Total	4,103.1	128,324,067	13.64	4,097.5	129,197,648	13.75	− 5.5	873,581	0.11	1
		Per patient	6.737	210,713		6.728	212,147		− 0.009	1,434
	TNBC	Total	8,538.1	431,548,773	20.44	8,534.6	434,122,775	20.77	− 3.4	2,574, 002	0.33	4
		Per patient	6.473	327,179		6.471	329, 130		− 0.003	1,951
	Breast Cancer	Total							− 32.1	5,690,990
		Per patient							-0.003	561
Shift to advanced stages for the UDC patients	Luminal A	Total	20,987.5	179,522,108	2.80	20,972.4	184,012,139	2.88	− 15.2	4,490,031	0.08	5
		Per patient	3.831	32,765		3.828	33,585		− 0.003	819
	Luminal B	Total	10,378.9	103,245,734	4.04	10,371.2	105,063,870	4.13	− 7.7	1,818,135	0.09	2
		Per patient	3.788	37,681		3.785	38,344		− 0.003	664
	HER2+	Total	2,242.4	82,155,748	8.06	2,239.5	82,838,914	8.14	− 2.9	683,165	0.08	1
		Per patient	3.682	134,903		3.677	136,024		− 0.005	1,122
	TNBC	Total	4,764.4	319,433,158	10.81	4,753.3	321,443,662	11.03	− 11.1	2,010,504	0.22	3
		Per patient	3.612	242,178		3.604	243,703		− 0.008	1,524
	Breast Cancer	Total							-36.9	9,001,835
		Per patient							-0.004	887
Shift in radiotherapy costs during COVID-19	Luminal A	Total	20,987.5	179,522,108	2.80	20,982.7	180,740,027	2.88	− 4.8	1,217,920	0.08	2
		Per patient	3.831	32,765		3.830	32,988		− 0.001	222
	Luminal B	Total	10,378.9	103,245,734	4.04	10,375.0	104,136,793	4.09	− 3.9	891,059	0.05	1
		Per patient	3.788	37,681		3.786	38,006		− 0.001	325
	HER2+	Total	2,242.4	82,155,748	8.06	2,240.0	82,856,543	8.13	− 2.5	700,794	0.07	1
		Per patient	3.682	134,903		3.678	136,053		− 0.004	1,151
	TNBC	Total	4,764.4	319,433,158	10.81	4,754.9	321,413,901	11.01	− 9.5	1,980,743	0.20	2
		Per patient	3.612	242,178		3.605	243,680		− 0.007	1,502
	Breast Cancer	Total							-20.7	4,790,516
		Per patient							-0.002	472

Open in a new tab

HER2 + Human Epidermal Growth Factor Receptor 2–positive, QALY Quality-Adjusted Life Year, TNBC Triple-Negative Breast Cancer, UDC Undiagnosed Cancer

Applying increased transition probabilities through multipliers (Table 4) from the UDC state towards advanced stage at diagnosis (i.e., stage III and IV) across subtypes resulted in nearly 37 QALYs lost and €9 million in additional healthcare costs, almost a threefold increase compared to the base case (Table 6). Luminal A and B subtypes incurred €4.49 million and €1.82 million in costs, and lost 15.17 and 7.70 QALYs, respectively. TNBC costs increased by €371,181 (to €2.01 million), and HER2+ by €136,165 (to €683,165). This scenario projected 11 additional deaths—almost twice the number in the base case—mainly among luminal subtypes. Absolute mortality increases were highest in TNBC (+ 0.22 percentage points), followed by luminal A and HER2 + (+ 0.08 each) and luminal B (+ 0.09). These corresponded to relative mortality increases of 2.0% for TNBC (from 10.81 to 11.03%), 2.9% for luminal A (from 2.80 to 2.88%), 2.2% for HER2 + (from 8.06 to 8.14%), and 2.2% for luminal B (from 4.04 to 4.13%).

When assuming that 8.2% of patients received a 5-fraction radiotherapy regimen [14]—costlier than the standard 15-fraction course—during the COVID-19 period, and that this practice persisted over five years, the model projected an increase of €1.60 million (to €4.79 million) in incremental breast cancer costs across all subtypes. This translated into a 2.5-fold increase for luminal A, 1.7-fold for luminal B, 1.3-fold for HER2+, and 1.2-fold for TNBC.

Probabilistic sensitivity analysis (PSA)

Scatterplots of the 1,000 PSA iterations were generated for each subtype to illustrate the distribution of incremental QALYs and healthcare costs (Figs. 4, 5, 6, 7). Table 8 presents the mean outcomes from the PSA by molecular subtype, including incremental QALYs and healthcare costs, as well as the corresponding 95% CIs.

Fig. 4 — Incremental QALYs (A) and healthcare costs (B) for the total luminal A cohort: Results from 1,000 PSA iterations. QALY: Quality-Adjusted Life Year; 95% CI: 95% confidence interval. Note. The red solid line represents the mean value across simulations. Black dashed lines indicate the 95% CIs. The shaded grey area marks the CI range. Percentages refer to the proportion of simulations showing gains vs. losses (A) or cost increases vs. savings (B)

Fig. 5 — Incremental QALYs (A) and healthcare costs (B) for the total luminal B cohort: Results from 1,000 PSA iterations. QALY: Quality-Adjusted Life Year; 95% CI: 95% confidence interval. Note. The red solid line represents the mean value across simulations. Black dashed lines indicate the 95% CIs. The shaded grey area marks the CI range. Percentages refer to the proportion of simulations showing gains vs. losses (A) or cost increases vs. savings (B)

Fig. 6 — Incremental QALYs (A) and healthcare costs (B) for the total HER2 + cohort: Results from 1,000 PSA iterations. QALY: Quality-Adjusted Life Year; 95% CI: 95% confidence interval. Note. The red solid line represents the mean value across simulations. Black dashed lines indicate the 95% CIs. The shaded grey area marks the CI range. Percentages refer to the proportion of simulations showing gains vs. losses (A) or cost increases vs. savings (B)

Fig. 7 — Incremental QALYs (A) and healthcare costs (B) for the total TNBC cohort: Results from 1,000 PSA iterations.ALY: Quality-Adjusted Life Year; 95% CI: 95% confidence interval. Note. The red solid line represents the mean value across simulations. Black dashed lines indicate the 95% CIs. The shaded grey area marks the CI range. Percentages refer to the proportion of simulations showing gains vs. losses (A) or cost increases vs. savings (B)

Table 8.

Probabilistic sensitivity analysis: QALY and cost changes by breast cancer subtype under care disruption

Molecular Subtype	Breast cancer cohort	Non-disrupted care (2017–2019)		Disrupted care (2020)		Incremental costs & QALYs (Disrupted care vs. non-disrupted care)
Molecular Subtype	Breast cancer cohort	QALYs	Cost (€)	QALYs	Cost (€)	∆QALYs	∆Cost (€)
Luminal A	Total	20,991.0	179,335,236	21,001.1	179,752,285	− 15.2	417,049
	95% CI	[20,962.2;21,019.8]	[178,767,318; 179,903,154)	[20,972.8; 21,029.3]	[179,187,369; 180,317,201]	[− 1,255.4; 1,249.99]	[− 26,790,493; 24,962,257]
Luminal B	Total	10,375.6	103,543,022	10,373.5	103,641,533	− 2.1	98,511
	95% CI	[10,361.8;10,389.4]	[103,113,677; 103,972,368]	[10,360.0; 10,386.9]	[103,219,807; 104,063,259]	[− 625.7; 604.0]	[− 19,523,679; 19,107,800]
HER2+	Total	2,243.4	82,244,443	2,242.3	82,474,170	− 1	229,727
	95% CI	[2,240.2; 2,246.6]	[81,876,267; 82,612,619]	[2,238.9; 2,245.7]	[82,126,409; 82,812,930]	[− 160.1; 154.1]	[− 16,775,684; 16,469,399]
Triple-Negative	Total	4,768.7	319,960,341	4,757.1	321,551,351	− 11.7	1,591,010
	95% CI	[4,762.5; 4,775.0]	[318,901,500; 321,019,181]	[4,754.1; 4,760.0]	[320,410,320; 322,692,381]	[− 213.0; 200.4]	[− 47,871,357; 53,229,055]

Open in a new tab

HER2 + Human Epidermal Growth Factor Receptor 2–positive, QALY Quality-Adjusted Life Year, 95% CI 95% confidence interval

Across all molecular subtypes, the PSA revealed wide uncertainty in incremental QALYs and costs. For luminal A, the 95% CI for incremental QALYs ranged from − 1255.36 to 1249.99, with a 51% probability of QALY loss, while incremental healthcare costs ranged from -€26,790,493 to €24,962,257, with a 52% probability of cost increase (Fig. 4). For luminal B, QALY changes ranged from − 625.65 to 604.00 (49% probability of loss) and costs from − €19,523,679 to €19,107,800 (50% probability of increase) (Fig. 5). For HER2+, QALYs ranged from − 160.11 to 154.10 (50% probability of loss) and costs from − €16,775,684 to €16,469,399 (52% probability of increase) (Fig. 6). For TNBC, QALYs ranged from − 213.03 to 200.42 (54% probability of loss) and costs from − €47,871,357 to €53,229,055 (51% probability of increase) (Fig. 7). These ranges illustrate that, while point estimates align with base-case findings (see Table 6), the CIs encompass both potential gains and losses in QALYs as well as cost savings and increases.

Discussion

This modelling exercise projected that breast cancer diagnostic delays between March and June 2020 in Belgium had only a modest health and economic impact over five years. Among 10,147 patients, the 135 undiagnosed cases resulted in an estimated 21 QALYs lost and six additional deaths, with wide uncertainty intervals including zero around the deterministic estimate. This averages to 0.002 QALYs per patient—representing less than one day in full health and falling far below the MCID [59]. Additionally, the model projected €3.2 million in extra healthcare costs, representing 1% of Belgian annual breast cancer expenditure. From a public health perspective, the disruption’s impact was limited and smaller than expected, despite early reports and widespread concern [60–62] during the first pandemic months.

Diagnostic delays affect all breast cancers, but their impact depends on tumour biology: missing an aggressive subtype such as TNBC or HER2 + carries greater risk than delaying diagnosis of luminal-like cancers [25]. In line with this rationale, our model applied clinically supported but hypothesised progression assumptions, which projected heterogeneity by subtype: TNBC and HER2 + together accounted for nearly 70% of excess healthcare costs and 60% of QALY losses.

Extending the time horizon to 10 years yielded similarly limited impacts, with over 32 QALYs lost—equivalent to around one day of full health per patient—and incremental healthcare costs representing < 2% (€5.7 million) of Belgian annual breast cancer expenditure. The relatively larger incremental QALY and cost impacts for luminal-like subtypes reflect their longer recurrence trajectories [27, 56], whereas most adverse outcomes for HER2 + and TNBC, which recur earlier [27, 56], were already captured within the five-year horizon. Even when assuming higher probabilities of UDC being diagnosed at advanced stages across all subtypes, the projected 5-year impact remained modest, with 37 QALYs lost (under two days in full health per patient) and €9 million in incremental costs (< 3% of annual breast cancer expenditure).

Although the deterministic base-case analysis projected modest incremental effects, the PSA revealed that these estimates are highly uncertain. Across all molecular subtypes, the 95% CIs spanned both positive and negative values, with probabilities of QALY loss or cost increase close to 50%. This indicates that the base-case results represent only one possible outcome within a wide distribution: small changes in key parameters could plausibly result in either greater harm or no adverse effect at all. Much of this uncertainty stems from transition probabilities from UDC to stage-specific diagnosis and from utility estimates, particularly for TNBC and HER2 + where data were limited. Consistent with this, the OSA identified TNBC inputs—with especially wide 95% CI—as having the greatest influence on overall results. The robustness of conclusions therefore hinges on the precision of such parameters, underscoring the need for higher-quality, subtype-specific data to reduce uncertainty.

The limited incremental QALYs and healthcare costs, despite wide uncertainty, are likely explained by the small proportion of UDC (1.33%) reported at the end of 2020, which were then modelled over a five-year period. Indeed, even though a larger proportion of patients were undiagnosed during the first few months of the pandemic [14], the majority was recovered by year-end. This is linked to Belgium’s screening programmes that resumed in June 2020 and ‘do-not-delay-care’ campaigns [63], proactive primary-care outreach [64] and teleconsultation policies [65], which supported timely symptomatic presentation. Consistent with this, a Belgian population-level observational study [14] found no evidence of a national stage shift, tumour size increase, or nodal progression in 2020, with DCIS diagnoses—likely most of the missing cases—rebounding by late 2020 and screening volumes recovering in 2021.

Our study is not the first to examine the potential health and economic consequences of COVID-19–related diagnostic delays in breast cancer. Sud et al. [66] modelled delays in the UK’s 2-week-wait referral pathway applying per-day hazard ratios for treatment delay uniformly to all stage I–III cases diagnosed through this route, which projected 733 excess breast cancer deaths and 15,339 life-years lost annually in England from a three-month delay. Maringe et al., [20] using linked NHS registry and hospital data in a stage-shift model, estimated 281–344 additional breast cancer deaths over five years in England due to pandemic-related diagnostic delays. Gheorghe et al. [24] then translated these mortality projections into 4,100 QALYs lost and £23.2 million in productivity losses. Degeling et al. [67], applying an Australian stage-shift model, projected 20–64 life-years lost and AU$2.0–7.7 million in extra costs for breast cancer under 3–6 month delay scenarios.

Differences between our findings and those of previous modelling studies can be explained by several factors. First, population sizes and healthcare systems differ between Belgium, the UK, and Australia, likely influencing the scale of the impact. Belgium’s smaller population, dense hospital network, and high acute bed capacity [68] supported resilience during the first wave, while rapid measures—such as reimbursed teleconsultations [69], proactive primary care outreach, adapted cancer guidelines [70], and hypofractionated radiotherapy [14]—helped sustain continuity of diagnosis and treatment. Importantly, breast cancer screening resumed in June 2020 [14], earlier than in parts of the UK [71, 72], which reduced the period of diagnostic delay. Coupled with strong inter-hospital collaboration [73] and national ‘do-not-delay-care’ campaigns, these factors limited the proportion of undiagnosed cases that progressed to later stages. This helps explain why our model projected smaller QALY losses, lower additional healthcare costs, and fewer excess deaths compared with modelling studies from the UK and Australia. In those regions, screening interruptions persisted longer and system capacity was more constrained. Second, prior models varied substantially in design: Sud et al. [66] applied per-day hazards of delay across all stage I–III cancers diagnosed via the 2-week-wait pathway; Maringe et al. [20] modelled large shifts from routine or screening routes to emergency presentations; Gheorghe et al. [24] translated Maringe et al. [20] excess deaths into QALY and productivity losses using a societal human-capital approach; and Degeling et al. [67] applied an inverse stage-shift method, conservatively modelling stage I to II progression under uniform 3–6 month delays. By contrast, our subtype-specific Markov model compared two Belgian cohorts: the observed 2020 breast cancer population (with 1.33% UDC) and a projected ‘no disruption’ cohort based on 2017–2019 trends, with UDC progression under diagnostic delay hypothesised to stage-specific diagnoses according to molecular subtype. Third, the duration and intensity of disruption differed markedly: Belgian screening was suspended for approximately three months with rapid recovery, whereas UK scenarios assumed screening disruptions lasting up to 12 months [66]. Finally, our analysis adopted a healthcare payer perspective, excluded DCIS and unknown-stage cases, and assumed constant treatment trajectories. These choices, combined with Belgium’s mitigation measures and the relatively rapid resumption of screening [14]—in contrast to Scotland [72] and Wales [71] for instance, where screening remained suspended until August 2020—may help explain the smaller health and economic impacts observed.

While this study focuses on the incremental healthcare costs (€3.2 million) and QALY losses (21 QALYs) associated with delayed breast cancer diagnosis during the first pandemic wave, it is important to recognise that the screening suspension also likely generated short-term reductions in resource use and patient burden. Some of the financial reduction was a deferral of expenditure, as the catch-up phase later in 2020 reinstated most screening costs [14]. However, genuine savings and avoided QALY losses likely arose from the temporary decline in false-positive work-ups and the avoidance of overdiagnosis of indolent tumours [74].

From a societal perspective, short-term economic effects may have included fewer lost workdays associated with screening attendance and follow-up visits, although potential longer-term productivity losses due to delayed diagnoses cannot be excluded. These effects could not be quantified, as no data exist on how temporary screening pauses affect absenteeism or anxiety at the population level. The implications for anxiety are more complex: while women may have avoided distress associated with false-positive recalls [75], this may have been offset by short-term pandemic-related stress and uncertainty surrounding postponed care.

While these aspects were not included in our healthcare payer model, they indicate that the broader societal impact of the screening disruption may be more nuanced than direct medical costs alone suggest, and must be weighed against the burden of additional advanced-stage aggressive cancers.

Strengths and limitations

The primary strength of this study lies in its use of a Markov model that integrates Belgian observational data on breast cancer incidence by clinical stage. To our knowledge, this is the first subtype-specific Markov model assessing the health and economic impact of COVID-19–related breast cancer diagnostic delays in Belgium. An additional strength is the explicit stratification by molecular subtype, a level of clinical detail rarely included in international COVID-19 modelling studies [20, 24, 66, 67, 76–78]. By accounting for tumour heterogeneity rather than treating breast cancer as a single entity, the model can reveal differential impacts across subtypes [12, 25]—particularly for more aggressive forms such as TNBC, which are highly sensitive to even short diagnostic delays. Despite uncertainties in input parameters, robust sensitivity analyses and expert consultations strengthened the model’s credibility and ensured that limitations were explicitly addressed. The framework is adaptable to other countries, providing a versatile policy tool [28]. Several limitations should be noted. While stage-specific projections from the BCR enabled estimation of UDC by stage in 2020, data were not subtype-specific. Applying the overall stage distribution uniformly across subtypes may overestimate the contribution of aggressive cancers, which are less likely to remain undiagnosed. Conversely, it may underestimate their impact, as missed cases would more plausibly present at later stages. A related limitation is that the model assumed transition probabilities from UDC to stage-specific diagnosis across molecular subtypes. The registry data did not allow us to distinguish diagnoses arising from resumed screening versus symptomatic presentations. National data, however, indicate that the recovery in late 2020 was driven by early-stage tumours in the screening age group [14], suggesting that screening activity primarily accounted for the catch-up. We therefore hypothesised that luminal-like subtypes would be diagnosed mostly at early stages, while aggressive subtypes would be more likely to present at advanced stages, reflecting symptomatic interval cancers. Some input parameters—such as transition probabilities from UDC to stage-specific diagnosis by subtype and selected disutilities (Table 4)—were not empirically available and were instead specified as plausible assumptions informed by clinical evidence. While this approach enabled model implementation, it inevitably introduces additional uncertainty in the incremental outcomes, as reflected in the PSA. Simplifying treatment trajectories (e.g., equating costs and utilities between luminal subtypes, not accounting for the modest increase in NAT use for early-stage disease [14]) may have attenuated subgroup-specific differences, producing more moderate aggregate results than would be expected if full clinical heterogeneity were incorporated. While age is a key prognostic factor and BCR data are available by age group, the absence of stratification by subtype prevented the model from accounting for age-related differences in progression, treatment response, or survival within subtypes—for instance, potentially larger QALY and healthcare cost impacts among younger women with aggressive cancers, and smaller impacts among older women due to higher competing mortality and less intensive treatment. The model assumed identical input parameters for disrupted and non-disrupted care, except for UDC and related transition probabilities. A third scenario incorporating Belgian data on radiotherapy reimbursement [79] (5-fraction vs. 15-fraction) showed that altering a single parameter could change incremental costs. Other adaptations—telemedicine [65], early discharge [80], and organizational changes—also likely affected costs and outcomes in both directions, so cumulative unmodelled effects may have differed from our estimates. Finally, certain clinical aspects were not modelled to avoid added complexity—for example, recurrence risks vary by stage and subtype, but LRR within stages I–III was not modelled separately.

Policy implications and future research

Although concerns during the initial pandemic suggested severe consequences, our analysis and BCR data indicate that the early disruption of breast cancer diagnosis had a modest five-year impact that, while growing slightly when projected over a 10-year horizon, remained limited overall. This likely reflects the rapid recovery of screening, proactive GP outreach, and targeted “do-not-delay-care” campaigns, which helped avoid a sustained stage shift. For policy, these findings highlight the need to protect continuity of diagnostic services during system shocks, minimise backlogs once services resume, and ensure vigilance for aggressive tumours such as TNBC and HER2+.

The modest aggregate impact observed following the short-term suspension of breast cancer screening should be interpreted with caution and does not, in itself, provide evidence to support changes to existing screening policies. While these findings primarily reflect the resilience of the healthcare system and timely recovery measures, they are also consistent with hypotheses proposed in the literature suggesting that brief interruptions may preferentially reduce the detection of indolent tumours, with limited short-term consequences for aggressive disease. Importantly, this interpretation is exploratory and hypothesis-generating, given the short duration of the disruption and the substantial uncertainty surrounding the estimates. Any consideration of optimising screening strategies, including approaches that account for tumour aggressiveness, would therefore require stronger empirical evidence from dedicated studies and improved data infrastructure. As noted in the limitations, the BCR does not routinely report incidence by molecular subtype, necessitating estimation in this study; systematic capture of subtype-specific incidence would be essential for future evaluations of screening efficiency.

Multi-state modelling is increasingly used in health-economic research because it can represent disease progression and time-dependent transitions more realistically than conventional cohort models (e.g., partitioned survival models, 3-state Markov models). In this study, it allowed us to explicitly model transitions from an ‘undiagnosed’ state to specific clinical stages and to quantify the downstream effects of diagnostic delay on recurrence and mortality in a detailed way. This flexible structure also supported scenario analyses—such as extending the time horizon, increasing transition probabilities from UDC towards advanced-stage disease across molecular subtypes, and modelling shifts in radiotherapy costs during COVID-19—making it a useful framework for evaluating health system resilience.

Future work should build on these insights by maintaining registry-based monitoring to capture possible longer-term impacts, particularly in aggressive subtypes, and by incorporating outcomes such as recurrence, overall survival, and HRQoL. An important next step will be for cancer registries to routinely provide subtype-specific incidence and stage data, enabling more granular assessment of differential impacts and supporting robust, policy-relevant modelling. It will also be important to evaluate the cumulative impact of multiple pandemic waves and to assess the effects of treatment adaptations (e.g., telemedicine, altered radiotherapy regimens, early discharge) that may have influenced QALY and healthcare cost changes. Finally, this study illustrates the value of rapid modelling exercises as a complement to slower-to-emerge observational data, providing policymakers with timely evidence to guide mitigation strategies and strengthen preparedness for future health system disruptions.

Conclusion

This study shows that breast cancer diagnostic delays during Belgium’s first COVID-19 wave were associated with modest five-year health and economic consequences, estimated at 21 QALYs lost, six additional deaths, and €3.2 million in additional healthcare costs (< 1% of annual expenditure). These smaller-than-expected impacts are reassuring and likely reflect Belgium’s rapid recovery of screening services and mitigation measures, which limited sustained stage shifts. Subtype-specific modelling nevertheless indicated that aggressive cancers, such as TNBC and HER2+, are more sensitive to diagnostic delays than luminal-like subtypes. However, projections were strongly assumption-driven and characterised by substantial uncertainty, underscoring the need for higher-quality, subtype-specific data. Continued monitoring of recurrence, survival, and HRQoL is therefore essential to capture potential longer-term effects. Ultimately, generating robust subtype-specific evidence will be crucial to inform future research on whether more tailored screening and diagnostic strategies could improve resilience during systemic healthcare disruptions.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary Material 1^{(5.6MB, xlsx)}

Supplementary Material 2^{(110.3KB, docx)}

Acknowledgements

The authors sincerely thank Dr. E.N. for her valuable clinical input on treatment protocols and cost estimations, which was instrumental in informing the healthcare cost calculations. Appreciation is also extended to all collaborators who contributed to the development of this work.

Abbreviations

AT: Adjuvant Therapy
BCPI: Belgian Centre for Pharmacotherapeutic Information
BCR: Belgian Cancer Registry
BCS: Breast-Conserving Surgery
CDK4/6: Cyclin-Dependent Kinases 4 and 6
COVID-19: Coronavirus Disease 2019
D&T: Diagnosis and Treatment
DCIS: Ductal Carcinoma In Situ
ER: Estrogen Receptor
ESMO: European Society for Medical Oncology
FU: Follow-Up
HER2: Human Epidermal Growth Factor Receptor 2
HRQoL: Health-Related Quality of Life
LRR: Locoregional Recurrence
MCID: Minimal Clinically Important Difference
MST: Mastectomy
NIHDI: National Institute for Health and Disability Insurance
OSA: One-Way Sensitivity Analysis
PR: Progesterone Receptor
PSA: Probabilistic Sensitivity Analysis
QALY: Quality-Adjusted Life Year
RT: Radiotherapy
SE: Standard Error
TNBC: Triple-Negative Breast Cancer
UDC: Undiagnosed Cancer
VBA: Visual Basic for Applications

Author contributions

YK conceived and designed the study, developed the Markov model, collected data inputs, performed the analyses, and drafted the manuscript. NV supervised the economic modelling, and critically reviewed the manuscript. CM contributed to the study design, provided clinical expertise, and validated treatment pathways. KV provided input on epidemiological aspects and critically reviewed the manuscript. SG critically reviewed the manuscript. BD critically reviewed the manuscript. FV and HP ensured oncological accuracy in the model and critically reviewed the manuscript. DS supervised the project, provided methodological guidance, and critically reviewed the manuscript.

Funding

This work was supported by Belgian Federal Science Policy Office, B2/202/P3/HELICON. The funder had no role in study design, analysis, interpretation, or manuscript writing.

Availability of data and materials

The Excel-based Markov model developed and analysed during the current study is provided as supplementary material with this article. In addition, a Word technical appendix (Additional file 1) is provided, containing detailed descriptions of the model structure, transition probabilities, utilities, healthcare costs, and assumptions used in the analysis.

Declarations

Ethics approval and consent to participate

This study was based on aggregated, de-identified registry data from the Belgian Cancer Registry (BCR), which exempted it from ethical review.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Footnotes

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

1.Hofmarcher T, Lindgren P, Wilking N, Jönsson B. The cost of cancer in Europe 2018. Eur J Cancer. 2020;129:41–9. [DOI] [PubMed] [Google Scholar]
2.JRC. Cancer care in times of COVID-19: lessons for future pandemics. 2023; Disponible sur: https://joint-research-centre.ec.europa.eu/jrc-news-and-updates/cancer-care-times-covid-19-lessons-future-pandemics-2023-02-28_en
3.WHO. Cancer. 2022; Disponible sur: https://www.who.int/news-room/fact-sheets/detail/cancer
4.Gorasso V, Vandevijvere S, Van der Heyden J, Pelgrims I, Hilderink H, Nusselder W, et al. The incremental healthcare cost associated with cancer in Belgium: a registry‐based data analysis. Cancer Med. 2024;13(3):e6659. [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Hanly P, Ortega-Ortega M, Soerjomataram I. Cancer premature mortality costs in Europe in 2020: a comparison of the human capital approach and the friction cost approach. Curr Oncol. 2022;29(5):3552–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
6.IARC. Cancer today - Age-Standardized Rate (World) per 100 000, Incidence and Mortality, Both sexes, in 2022 [Internet]. 2024. Disponible sur: https://gco.iarc.who.int/today/en/dataviz/bars?mode=population%26;group_populations=0%26;types=0_1%26;sort_by=value1%26;cancers=20%26;populations=%26;nb_items=-1%26;sexes=2
7.AIHTA. Oncological Breast Cancer Care in Selected European Countries. 2024; Disponible sur: https://eprints.aihta.at/1545/1/HTA-Projektbericht_Nr.162.pdf
8.WHO. Overview of Public Health and Social Measures in the context of COVID-19 [Internet]. 2020 [cité 2 juill 2025]. Disponible sur: https://www.who.int/publications/i/item/overview-of-public-health-and-social-measures-in-the-context-of-covid-19
9.World Health Organization. Pulse survey on continuity of essential health services during the COVID-19 pandemic: interim report, 27 August 2020. 2020 [cité 24 avr 2025]; Disponible sur: https://www.who.int/publications/i/item/WHO-2019-nCoV-EHS_continuity-survey-2020.1
10.Vanni G, Materazzo M, Pellicciaro M, Ingallinella S, Rho M, Santori F, et al. Breast cancer and COVID-19: the effect of fear on patients’ decision-making process. In Vivo. 2020;34(3 Suppl):1651. [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Abdel-Razeq H, Mansour A, Edaily S, Dayyat A. Delays in initiating anti-cancer therapy for early-stage breast cancer—how slow can we Go? J Clin Med. 2023;12(13):4502. [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Caplan L. Delay in breast cancer: implications for stage at diagnosis and survival. Front Public Health. 2014;2:87. [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Hanna TP, King WD, Thibodeau S, Jalink M, Paulin GA, Harvey-Jones E, et al. Mortality due to cancer treatment delay: systematic review and meta-analysis. BMJ. 2020;371:m4087. [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Peacock HM, van Walle L, Silversmit G, Neven P, Han SN, Van Damme N. Breast cancer incidence, stage distribution, and treatment shifts during the 2020 COVID-19 pandemic: a nationwide population-level study. Arch Public Health. 2024;82:66. [DOI] [PMC free article] [PubMed] [Google Scholar]
15.ESMO. ESMO Clinical Practice Guideline: Early Breast Cancer [Internet]. 2024 [cité 4 juill 2025]. Disponible sur: https://www.esmo.org/guidelines/esmo-clinical-practice-guideline-early-breast-cancer
16.Gennari A, André F, Barrios CH, Cortés J, de Azambuja E, DeMichele A, et al. ESMO Clinical Practice Guideline for the diagnosis, staging and treatment of patients with metastatic breast cancer☆. Ann Oncol. 2021;32(12):1475–95. [DOI] [PubMed] [Google Scholar]
17.Orrantia-Borunda E, Anchondo-Nuñez P, Acuña-Aguilar LE, Gómez-Valles FO, Ramírez-Valdespino CA. Subtypes of Breast Cancer. In: Mayrovitz HN, éditeur. Breast Cancer [Internet]. Brisbane (AU): Exon Publications; 2022 [cité 25 janv 2025]. Disponible sur: http://www.ncbi.nlm.nih.gov/books/NBK583808/ [PubMed]
18.van Walle L. Incidence of breast cancer subtypes in Belgium: a population-based study. BJMO [Internet]. 2020 [cité 16 avr 2025]; Disponible sur: https://www.bjmo.be/journal-article/incidence-of-breast-cancer-subtypes-in-belgium-a-population-based-study/
19.Resende CAA, Fernandes Cruz HM, Costa e Silva M, Paes RD, Dienstmann R, Barrios CHE, et al. Impact of the COVID-19 pandemic on cancer staging: an analysis of patients with breast cancer from a community practice in Brazil. JCO Glob Oncol. 2022;8:e2200289. [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Maringe C, Spicer J, Morris M, Purushotham A, Nolte E, Sullivan R, et al. The impact of the COVID-19 pandemic on cancer deaths due to delays in diagnosis in England, UK: a national, population-based, modelling study. Lancet Oncol août. 2020;21(8):1023–34. [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Alagoz O, Lowry KP, Kurian AW, Mandelblatt JS, Ergun MA, Huang H, et al. Impact of the COVID-19 pandemic on breast cancer mortality in the US: estimates from collaborative simulation modeling. JNCI J Natl Cancer Inst. 2021;113(11):1484–94. [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Yong JH, Mainprize JG, Yaffe MJ, Ruan Y, Poirier AE, Coldman A, et al. The impact of episodic screening interruption: COVID-19 and population-based cancer screening in Canada. J Med Screen juin. 2021;28(2):100–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Dul M, Grzeszczyk MK, Nojszewska E, Sitek A. Estimation of the Impact of COVID-19 Pandemic Lockdowns on Breast Cancer Deaths and Costs in Poland using Markovian Monte Carlo Simulation [Internet]. arXiv; 2023 [cité 4 déc 2024]. Disponible sur: http://arxiv.org/abs/2305.00908
24.Gheorghe A, Maringe C, Spice J, Purushotham A, Chalkidou K, Rachet B, et al. Economic impact of avoidable cancer deaths caused by diagnostic delay during the COVID-19 pandemic: A national population-based modelling study in England, UK. Eur J Cancer. 2021;152:233–42. [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Alsalamah RA, Alsalamah DRA. Critical Diagnostic Delay Thresholds in Breast Cancer: A Molecular Subtype-Based Causal Analysis From Saudi Arabia. Cureus [Internet]. 24 mars 2025 [cité 29 mars 2025];17(3). Disponible sur: https://www.cureus.com/articles/351382-critical-diagnostic-delay-thresholds-in-breast-cancer-a-molecular-subtype-based-causal-analysis-from-saudi-arabia [DOI] [PMC free article] [PubMed]
26.KCE. BELGIAN GUIDELINES FOR ECONOMIC EVALUATIONS AND BUDGET IMPACT ANALYSES: THIRD EDITION [Internet]. 2025. Disponible sur: https://kce.fgov.be/sites/default/files/2025-05/KCE400_Method_guidelines_economic_evaluations.pdf
27.van Maaren MC, de Munck L, Strobbe LJA, Sonke GS, Westenend PJ, Smidt ML, et al. Ten-year recurrence rates for breast cancer subtypes in the Netherlands: a large population-based study. Int J Cancer. 2019;144(2):263–72. [DOI] [PubMed] [Google Scholar]
28.Khan Y, Verhaeghe N, Pauw RD, Devleesschauwer B, Gadeyne S, Gorasso V, et al. Evaluating the health and health economic impact of the COVID-19 pandemic on delayed cancer care in Belgium: a Markov model study protocol. PLoS ONE. 2023;18(10):e0288777. [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Sonnenberg FA, Beck JR. Markov models in medical decision making: a practical guide. Med Decis Making. 1993;13(4):322–38. [DOI] [PubMed] [Google Scholar]
30.Belgian Cancer Registry. Mission. 2024; Disponible sur: https://kankerregister.org/en/mission
31.Lobo-Martins S, Arecco L, Cabral TP, Agostinetto E, Dauccia C, Franzoi MA, et al. Extended adjuvant endocrine therapy in early breast cancer: finding the individual balance. ESMO Open [Internet]. 1 mai 2025 [cité 8 juill 2025];10(5). Disponible sur: https://www.esmoopen.com/article/S2059-7029(25)00926-3/fulltext [DOI] [PMC free article] [PubMed]
32.Intrieri T, Manneschi G, Caldarella A. 10-year survival in female breast cancer patients according to ER, PR and HER2 expression: a cancer registry population-based analysis. J Cancer Res Clin Oncol. 2022;149(8):4489–96. [DOI] [PMC free article] [PubMed] [Google Scholar]
33.RIVIZ. Nomensoft: Nomenclatuur en pseudonomenclatuur van de geneeskundige verstrekkingen [Internet]. 2025. Disponible sur: https://webappsa.riziv-inami.fgov.be/Nomen/fr/search
34.BCFI. Belgisch Centrum voor Farmacotherapeutische Informatie [Internet]. 2025. Disponible sur: https://www.bcfi.be/nl/
35.Statbel. Health Index. 2025; Disponible sur: https://statbel.fgov.be/en/themes/consumer-prices/health-index
36.Van Wilder L, Charafeddine R, Beutels P, Bruyndonckx R, Cleemput I, Demarest S, et al. Belgian population norms for the EQ-5D-5L, 2018. Qual Life Res Int J Qual Life Asp Treat Care Rehabil. 2022;31(2):527–37. [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Kregting LM, Vrancken Peeters NJMC, Clarijs ME, Koppert LB, Korfage IJ, van Ravesteyn NT. Health utility values of breast cancer treatments and the impact of varying quality of life assumptions on cost-effectiveness. Int J Cancer. 2024;155(1):117–27. [DOI] [PubMed] [Google Scholar]
38.Torres S, Bayoumi AM, Abrahao ABK, Trudeau M, Pritchard KI, Li CN, et al. Implementing routine collection of EQ-5D-5L in a breast cancer outpatient clinic. PLoS ONE. 2024;19(8):e0307225. [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Jacobs DHM, Horeweg N, Straver M, Roeloffzen EMA, Speijer G, Merkus J, et al. Health-related quality of life of breast cancer patients after accelerated partial breast irradiation using intraoperative or external beam radiotherapy technique. The Breast. 2019;46:32–9. [DOI] [PubMed] [Google Scholar]
40.Sundaresan P, Sullivan L, Pendlebury S, Kirby A, Rodger A, Joseph D, et al. Patients’ perceptions of health-related quality of life during and after adjuvant radiotherapy for T1N0M0 breast cancer. Clin Oncol. 2015;27(1):9–15. [DOI] [PubMed] [Google Scholar]
41.Khoirunnisa SM, Suryanegara FDA, Setiawan D, Postma MJ. Health-related quality of life in Her2-positive early breast cancer woman using trastuzumab: a systematic review and meta-analysis. Front Pharmacol. 2023;14:1090326. [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Rugo H, Brammer M, Zhang F, Lalla D. Effect of trastuzumab on health-related quality of life in patients with HER2-positive metastatic breast cancer: data from three clinical trials. Clin Breast Cancer. 2010;10(4):288–93. [DOI] [PubMed] [Google Scholar]
43.Zhao Y, Chen L, Zheng X, Shi Y. Quality of life in patients with breast cancer with neoadjuvant chemotherapy: a systematic review. BMJ Open. 2022;12(11):e061967. [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Liu M, Goldberg J, Norton L, Robson ME, Zhi I. Immunotherapy related patient-reported outcomes from five randomized controlled trials in patients with triple negative breast cancer: a systematic review. J Clin Oncol. 2024. 10.1200/JCO.2024.42.16_suppl.e23167.39693584 [Google Scholar]
45.Lidgren M, Wilking N, Jönsson B, Rehnberg C. Health related quality of life in different states of breast cancer. Qual Life Res Int J Qual Life Asp Treat Care Rehabil. 2007;16(6):1073–81. [DOI] [PubMed] [Google Scholar]
46.Eggersmann TK, Degenhardt T, Gluz O, Wuerstlein R, Harbeck N. CDK4/6 inhibitors expand the therapeutic options in breast cancer: palbociclib. Ribociclib Abemaciclib BioDrugs. 2019;33(2):125–35. [DOI] [PubMed] [Google Scholar]
47.Mason SR, Willson ML, Egger SJ, Beith J, Dear RF, Goodwin A. Platinum chemotherapy for early triple-negative breast cancer. The Breast [Internet]. 1 juin 2024 [cité 29 mars 2025];75. Disponible sur: https://www.thebreastonline.com/article/S0960-9776(24)00043-2/fulltext#fig3 [DOI] [PMC free article] [PubMed]
48.Di Leo A, Jerusalem G, Petruzelka L, Torres R, Bondarenko IN, Khasanov R, et al. Results of the CONFIRM Phase III Trial Comparing Fulvestrant 250 mg with Fulvestrant 500 mg in postmenopausal women with estrogen receptor-positive advanced breast cancer. J Clin Oncol. 2010;28(30):4594–600. [DOI] [PubMed] [Google Scholar]
49.Paracha N, Reyes A, Diéras V, Krop I, Pivot X, Urruticoechea A. Evaluating the clinical effectiveness and safety of various HER2-targeted regimens after prior taxane/trastuzumab in patients with previously treated, unresectable, or metastatic HER2-positive breast cancer: a systematic review and network meta-analysis. Breast Cancer Res Treat. 2020;180(3):597–609. [DOI] [PMC free article] [PubMed] [Google Scholar]
50.Bardia A, Hurvitz SA, Tolaney SM, Loirat D, Punie K, Oliveira M, et al. Sacituzumab govitecan in metastatic triple-negative breast cancer. N Engl J Med. 2021;384(16):1529–41. [DOI] [PubMed] [Google Scholar]
51.Hurvitz SA, Kim SB, Chung WP, Im SA, Park YH, Hegg R, et al. Trastuzumab deruxtecan versus trastuzumab emtansine in HER2-positive metastatic breast cancer patients with brain metastases from the randomized DESTINY-Breast03 trial. ESMO Open [Internet]. 1 mai 2024 [cité 29 mars 2025];9(5). Disponible sur: https://www.esmoopen.com/article/S2059-7029%2824%2900692-6/fulltext [DOI] [PMC free article] [PubMed]
52.Cortes J, Hudgens S, Twelves C, Perez EA, Awada A, Yelle L, et al. Health-related quality of life in patients with locally advanced or metastatic breast cancer treated with eribulin mesylate or capecitabine in an open-label randomized phase 3 trial. Breast Cancer Res Treat déc. 2015;154(3):509–20. [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Modi S, Jacot W, Yamashita T, Sohn J, Vidal M, Tokunaga E, et al. Trastuzumab deruxtecan in previously treated HER2-low advanced breast cancer. N Engl J Med. 2022;387(1):9–20. [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Mueller V, Wardley A, Paplomata E, Hamilton E, Zelnak A, Fehrenbacher L, et al. Preservation of quality of life in patients with human epidermal growth factor receptor 2-positive metastatic breast cancer treated with tucatinib or placebo when added to trastuzumab and capecitabine (HER2CLIMB trial). Eur J Cancer Oxf Engl. 1990;2021(153):223–33. [DOI] [PubMed] [Google Scholar]
55.Valerio MR, Spadaro P, Arcanà C, Borsellino N, Cipolla C, Vigneri P, et al. Oral vinorelbine and capecitabine as first-line therapy in metastatic breast cancer: a retrospective analysis. Future Sci OA. 2021;7(10):FSO750. [DOI] [PMC free article] [PubMed] [Google Scholar]
56.Pedersen RN, Esen BÖ, Mellemkjær L, Christiansen P, Ejlertsen B, Lash TL, et al. The incidence of breast cancer recurrence 10–32 years after primary diagnosis. J Natl Cancer Inst. 2022;114(3):391–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
57.Tokisawa H, Aruga T, Honda Y, Ishiba T, Yonekura R, Iwamoto N, et al. Abstract P4–07–25: Distant metastasis of breast cancer is triggered by changes in the dynamics of metastatic cells after removal of the primary lesion. Cancer Res. 2022;82(4_Supplement):P4-7–25. [Google Scholar]
58.RIZIV. Verkorte bestraling bij borstkankerpatiënten tijdens de COVID-19-pandemie [Internet]. 2020. Disponible sur: https://www.riziv.fgov.be/nl/thema-s/covid-19/verkorte-bestraling-bij-borstkankerpatienten-tijdens-de-covid-19-pandemie
59.Bourke S, Bennett B, Oluboyede Y, Li T, Longworth L, O’Sullivan SB, et al. Estimating the minimally important difference for the EQ-5D-5L and EORTC QLQ-C30 in cancer. Health Qual Life Outcomes. 2024;22(1):81. [DOI] [PMC free article] [PubMed] [Google Scholar]
60.Simão D, Sardinha M, Reis AF, Spencer AS, Luz R, Oliveira S, et al. What Has Changed During the COVID-19 Pandemic? - The Effect on an Academic Breast Department in Portugal. Eur J Breast Health [Internet]. 30 déc 2021 [cité 22 août 2025]; Disponible sur: https://eurjbreasthealth.com/articles/what-has-changed-during-the-covid-19-pandemic-the-effect-on-an-academic-breast-department-in-portugal/ejbh.galenos.2021.2021-11-1 [DOI] [PMC free article] [PubMed]
61.European Federation of Pharmaceutical Industries and Associations. The impact of COVID-19 on patient access to cancer care in Europe [Internet]. 2021. Disponible sur: https://www.efpia.eu/media/602636/every-day-counts-covid19-addendum.pdf
62.Peacock HM, Van Meensel M, Van Gool B, Silversmit G, Dekoninck K, Brierley JD, et al. Cancer incidence, stage shift and survival during the 2020 COVID-19 pandemic: a population-based study in Belgium. Int J Cancer. 2024. 10.1002/ijc.35001. [DOI] [PubMed] [Google Scholar]
63.Konusevska A. European Cancer Organisation. 2024 [cité 22 août 2025]. The Time to Act is Now. Action Report from Covid-19 & Cancer Workforce Special Network Meeting. Disponible sur: https://www.europeancancer.org/resources/publications/reports/the-time-to-act-is-now-action-report.html
64.Peacock HM, Tambuyzer T, Verdoodt F, Calay F, Poirel HA, Schutter HD, et al. Decline and incomplete recovery in cancer diagnoses during the COVID-19 pandemic in Belgium: a year-long, population-level analysis. ESMO Open. 2021. 10.1016/j.esmoop.2021.100197. [DOI] [PMC free article] [PubMed] [Google Scholar]
65.Gottlob A, Schmitt T, Frydensberg MS, Rosińska M, Leclercq V, Habimana K. Telemedicine in cancer care: lessons from COVID-19 and solutions for Europe. Eur J Public Health. 2025;35(1):35–41. [DOI] [PMC free article] [PubMed] [Google Scholar]
66.Sud A, Torr B, Jones ME, Broggio J, Scott S, Loveday C, et al. Effect of delays in the 2-week-wait cancer referral pathway during the COVID-19 pandemic on cancer survival in the UK: a modelling study. Lancet Oncol. 2020;21(8):1035–44. [DOI] [PMC free article] [PubMed] [Google Scholar]
67.Degeling K, Baxter NN, Emery J, Jenkins MA, Franchini F, Gibbs P, et al. An inverse stage-shift model to estimate the excess mortality and health economic impact of delayed access to cancer services due to the COVID-19 pandemic. Asia Pac J Clin Oncol. 2021;17(4):359–67. [DOI] [PMC free article] [PubMed] [Google Scholar]
68.OECD. Belgium: Country Health Profile 2023 [Internet]. 2023. Disponible sur: https://www.oecd.org/content/dam/oecd/en/publications/reports/2023/12/belgium-country-health-profile-2023_01d0f3f9/dd6df7bd-en.pdf
69.INAMI. Remboursement des consultations médicales à distance [Internet]. 2020. Disponible sur: https://www.inami.fgov.be/fr/themes/soins-de-sante-cout-et-remboursement/les-prestations-de-sante-que-vous-rembourse-votre-mutualite/consultations-medicales-a-distance
70.ESMO. ESMO management and treatment adapted recommendations in the COVID-19 era: Breast cancer [Internet]. 2020. Disponible sur: https://dam.esmo.org/image/upload/v1755611032/ESMO-guidelines-COVID19-breast-cancer-2020-04-15_cyxwsi.pdf [DOI] [PMC free article] [PubMed]
71.NHS Wales. Breast Test Wales - Annual Statistical Report - 2020–2021 [Internet]. 2024. Disponible sur: https://phw.nhs.wales/publications/publications1/breast-test-wales-annual-report-202021/
72.Scottish Government. Breast Cancer Screening to resume [Internet]. 2020. Disponible sur: https://www.gov.scot/news/breast-cancer-screening-to-resume/
73.Kenis I, Theys S, Hermie E, Foulon V, Van Hecke A. Impact of COVID-19 on the organization of cancer care in Belgium: lessons learned for the (post-)pandemic future. Int J Environ Res Public Health. 2022;19(19):12456. [DOI] [PMC free article] [PubMed] [Google Scholar]
74.Bosch G, Posso M, Louro J, Roman M, Porta M, Castells X, et al. Impact of the COVID-19 pandemic on breast cancer screening indicators in a Spanish population-based program: a cohort study. Elife. 2022;11:e77434. [DOI] [PMC free article] [PubMed] [Google Scholar]
75.Bolejko A, Hagell P, Wann-Hansson C, Zackrisson S. Prevalence, long-term development, and predictors of psychosocial consequences of false-positive mammography among women attending population-based screening. Cancer Epidemiol Biomarkers Prev. 2015;24(9):1388–97. [DOI] [PubMed] [Google Scholar]
76.Malagón T, Yong JHE, Tope P, Miller WHJr, Franco EL, McGill Task Force on the Impact of COVID-19 on Cancer Control and Care. Predicted long-term impact of COVID-19 pandemic-related care delays on cancer mortality in Canada. Int J Cancer. 2022;150(8):1244–54. [DOI] [PMC free article] [PubMed] [Google Scholar]
77.Zhang M, Tierney P, Brennan A, Murray D, Mullooly M, Bennett K. Modelling the impact of the COVID-19 pandemic on cancer stage migration and excess mortality in Ireland. Prev Med Rep. 2025;52:103020. [DOI] [PMC free article] [PubMed] [Google Scholar]
78.Figueroa JD, Gray E, Maeda Y, Hall PS, Mackean M, Elder K, et al. Modelling the Long-Term Effects of Covid-19 Cancer Services Disruption on Patient Outcome in Scotland [Internet]. medRxiv; 2021 [cité 26 août 2025]. p. 2021.01.17.21249993. Disponible sur: https://www.medrxiv.org/content/10.1101/2021.01.17.21249993v1
79.service public federal securite sociale. Arrêté Royal du 13/05/2020 arrete royal n° 20 portant des mesures temporaires dans la lutte contre la pandemie covid-19 et visant a assurer la continuite des soins en matiere d’assurance obligatoire soins de sante [Internet]. Moniteur Belge; 2020 mai [cité 11 juill 2025]. Disponible sur: https://etaamb.openjustice.be/fr/arrete-royal-du-13-mai-2020_n2020041295
80.Faes C, Abrams S, Van Beckhoven D, Meyfroidt G, Vlieghe E, Hens N, et al. Time between symptom onset, hospitalisation and recovery or death: statistical analysis of Belgian COVID-19 patients. Int J Environ Res Public Health. 2020;17(20):7560. [DOI] [PMC free article] [PubMed] [Google Scholar]
81.Mao JH, van Diest PJ, Perez-Losada J, Snijders AM. Revisiting the impact of age and molecular subtype on overall survival after radiotherapy in breast cancer patients. Sci Rep. 2017;7(1):12587. [DOI] [PMC free article] [PubMed] [Google Scholar]
82.Allen K, Lohrisch CA, Le D, Diocee RM, Speers C, Nichol A, et al. Survival following locoregional recurrence in breast cancer by clinical subtype. J Clin Oncol. 2021;39(15_suppl):543–543. [Google Scholar]
83.Hartkopf AD, Walter CB, Kolberg HC, Hadji P, Tesch H, Fasching PA, et al. Attrition in the first three therapy lines in patients with advanced breast cancer in the German real-world PRAEGNANT registry. Geburtshilfe Frauenheilkd mai. 2024;84(5):459–69. [DOI] [PMC free article] [PubMed] [Google Scholar]
84.Bardia A, Hu X, Dent R, Yonemori K, Barrios CH, O’Shaughnessy JA, et al. Trastuzumab deruxtecan after endocrine therapy in metastatic breast cancer. N Engl J Med. 2024;391(22):2110–22. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Material 1^{(5.6MB, xlsx)}

Supplementary Material 2^{(110.3KB, docx)}

PERMALINK

Subtype-specific health and economic impact of delayed breast cancer diagnosis during the early COVID-19 pandemic in Belgium: a Markov model analysis

Yasmine Khan

Nick Verhaeghe

Chris Monten

Katrien Vanthomme

Sylvie Gadeyne

Brecht Devleesschauwer

Freija Verdoodt

Hanna M Peacock

Delphine De Smedt

Abstract

Background

Methods

Results

Conclusion

Supplementary Information

Background

Table 1.

Methods

Study design

Table 2.

Fig. 1.

Table 4.

Markov model structure

Table 3.

Model validation

Data inputs

Epidemiological data

Transition probabilities

Healthcare costs

Table 5.

Utilities

Analysis

Scenario and sensitivity analyses

Results

Base case analysis

Table 6.

One-way sensitivity analysis (OSA)

Fig. 2.

Fig. 3.

The impact of varying transition probabilities and utilities on incremental QALYs

The impact of varying model inputs on incremental healthcare cost changes

Scenario analyses

Table 7.

Probabilistic sensitivity analysis (PSA)

Fig. 4.

Fig. 5.

Fig. 6.

Fig. 7.

Table 8.

Discussion

Strengths and limitations

Policy implications and future research

Conclusion

Supplementary Information

Acknowledgements

Abbreviations

Author contributions

Funding

Availability of data and materials

Declarations

Ethics approval and consent to participate

Consent for publication

Competing interests

Footnotes

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases