Impact of breast cancer care pathways and related symptoms on the return-to-work process: results from a population-based French cohort study (CONSTANCES)

The study sample was from the CONSTANCES cohort, a population-based prospective cohort study that included ~ 220,000 volunteers aged from 18 to 69 years at 21 health examination centers throughout France between 2012 and 2020 [23]. At inclusion, a health examination was performed, and self-report questionnaires were given to participants to collect sociodemographic, lifestyle, socio-professional data and medical history. The follow-up included self-report questionnaires (filled in at home) once per year and a health examination every 4 years [23]. The CONSTANCES cohort was linked to the national health insurance system database [24] (SNDS) to obtain additional data on BC diagnosis, BC treatments, other treatments reimbursed to the patients due to BC-related problems, such as anxiolytic/antidepressant and antalgic drugs, and statutory sick pay from January 2012 to December 2019. Women with breast cancer were identified in the SNDS from hospitalization records (principal or related diagnoses with two ICD-10 codes = C50—Malignant neoplasm of breast, D05—Carcinoma in situ of breast) and/or long-term chronic disease benefits (ICD-10 = C50, D05). The quality of the BC’s identification in the SNDS was previously tested using medical records for a subsample of 265 cases, resulting in a high positive predictive value (PPV = 92% 95% CI [88–95%]).

Analyses were restricted to women from the CONSTANCES cohort who were working at the time of BC diagnosis (from 2012 to 2018) and who gave their informed consent for SNDS data collection (Fig. 1). BC cases were predominantly diagnosed the year of their inclusion in the CONSTANCES cohort or the year before (n = 391/626, 62.5%).

RTW and time to RTW were estimated using data on sick leaves (SL) identified from the daily sick pay in the SNDS. A woman with BC was considered as having gone back to work when her daily sick pay was stopped. No distinction between partial- and full-time RTW was made in the statistical analysis. Women on therapeutic partial sick leave were considered to have gone back to work. Patients were considered to have a BC-linked SL if it lasted more than 21 days to avoid SL due to other occasional health issues. The time to RTW was calculated from the first day of the > 21-day-long SL to the last day of the sick pay in the 2 years after the BC diagnosis. Civil servants and self-employed women, for whom sick pay data are not available in the SNDS, were excluded from the statistical analysis on RTW and time to RTW (Fig. 1).

BC care pathways were identified using the sequence analysis method that allowed to consider the temporal order of successive BC treatments [16, 17, 25, 26]. Each woman was represented by a sequence of states over time that corresponded to different BC treatments she received each month, from the first month of BC treatment identified in the SNDS (t₀) to month 24 after treatment initiation (t₂₄). Seven states were included: the five BC treatment types identified in the SNDS (mastectomy, breast-conserving surgery, chemotherapy, radiotherapy, and breast reconstructive surgery), the chemotherapy and radiotherapy state (i.e., women who underwent alternatively these two treatments within the same month), and treatment-free months. Axillary lymph node (LN) dissection and hormone therapy also were recorded in the SNDS, but they were not included in the BC treatment sequences. Indeed, axillary LN dissection is usually performed during breast surgery, and hormone therapy is mainly used to prevent BC recurrence and is prescribed over a long period of time (5 and up to 10 years after diagnosis), usually when women are already back at work.

Then, the optimal matching algorithm [27] was used to assess the dissimilarity among the women’s sequence of treatments. Treatment sequences were clustered into homogeneous patterns by hierarchical clustering using the Ward criterion. The appropriate number of clusters was chosen on the basis of the dendrogram and the histogram of the gaps in inertia, and also according to the clinical recommendations for BC care and statistical constraints (i.e., enough patients in each cluster) (Additional file 1: Fig S1).

Sociodemographic characteristics (age, education level, household income, household composition, occupational category) were collected from the CONSTANCES self-report questionnaires filled in by the patients at inclusion or during the follow-up (in function of the BC diagnosis date). Information on BC treatments (breast surgery, axillary LN dissection, chemotherapy, radiotherapy, reconstructive surgery) and reimbursement of anxiolytic/antidepressant and antalgic drugs (used as a proxy of depressive symptoms and pain in the 2 years after BC diagnosis) were obtained monthly from the SNDS for the 2012–2019 period (Table 1).

BC care (BC pathways, antidepressant/anxiolytic and antalgic drug deliveries) and sociodemographic characteristics were described for all women who were working at the time of BC diagnosis (n = 626, Fig. 1).

The median follow-up time was estimated by the reverse Kaplan–Meier method [28].

Multivariate Cox models with time-dependent covariates were used to assess the influence of BC treatments on the probability to return to work. Time-dependent variables were: BC treatments (except for breast reconstruction due to the limited number of women in the cohort who received this treatment) and anxiolytic and antalgic drug deliveries (measured monthly). Sociodemographic variables, measured only at BC diagnosis, were considered as fixed variables and were selected from a limited number of variables available in the CONSTANCES questionnaire that were previously described as potential predictors of RTW and/or time to RTW [7, 8, 10, 14, 22]. For each outcome, two multivariate models were performed. In the first multivariate model (model A), each BC treatment was treated as an independent variable, whereas in the second model (model B), the BC care pathway patterns identified by sequence analysis were used instead of the individual BC treatments.

The proportional hazards assumption was tested for the fixed variables included in the Cox models using the Schoenfeld residuals analysis and the Hosmer–Lemeshow test.

Statistical significance was defined by a p value < 0.05. All statistical analyses were performed with the R statistical software, version 4.0.2, using the packages 'survival,' 'survminer' and TraMineR'[25].

Data for 89,522 of the 106,932 women included in the CONSTANCES cohort between 2012 and 2019 could be linked to SNDS database. In the group with matched data, 1,213 women received a diagnosis of BC between 2012 and 2018 among whom 626 were working at diagnosis time. SL data were available in the SNDS for 349/626 active women, and a SL > 21 days within the two years after BC diagnosis was found for 303 of these women (Fig. 1).

The median age at diagnosis was 49.1 years (IQR 44.0–54.3). Among the 626 active women at BC diagnosis, 80% underwent breast conservative surgery, 25.4% total mastectomy, 57.8% axillary LN dissection, 45.4% chemotherapy and 44.7% radiotherapy. Moreover, 68.4% of active women received hormone therapy, and 61% and 63.4% had at least one anxiolytic and antalgic drug delivery within the two years after BC diagnosis, respectively (Table 1).

BC care pathways in the two years after the first treatment were identified for the 626 women working at BC diagnosis. Five patterns were determined (Fig. 2): “Surgery” (S, n = 201) when treatment was limited to curative BC surgery; “Surgery and Radiotherapy” (SR, n = 145) when BC surgery was followed by radiotherapy only; “Surgery and Chemotherapy” (SC, n = 102) when BC surgery was followed by chemotherapy only; “Surgery, Chemotherapy, and Radiotherapy” (SCR, n = 107) when BC surgery was followed by chemotherapy and radiotherapy (mainly in this order); and “Surgery and Long Chemotherapy” (SLC, n = 71) when surgery was followed by a long-term chemotherapy. A part of BCS included in the SLC pattern also underwent an alternating of radiotherapy and chemotherapy sessions and treatment within their care pathways.

The median SL duration for the 303 women with SL > 21 days was 9.8 months (mean 11.6, IQR 3.7–15.1, Additional file 1: Table S1) and the median follow-up of the cohort was 22.6 months (Fig. 3). The RTW rate was 33.0% (95%CI 27.7–38.3), 55.8%, (95%CI 50.2–61.4) % and 69.3%, (95%CI 64.1–74.5), 73.2%, (95%CI 71.4–81.0), respectively, at 6 months, 12 months, 18 months and 24 month after BC diagnosis (Fig. 3). The time to RTW varied according to different BC care pathway patterns and increased with the complexity and duration of BC treatments (p < 0.001) (Fig. 3 and Additional file 1: Table S1).

Table 2 presents the associations between probability of RTW and BC treatments, anxiolytic and antalgic drug deliveries, and sociodemographic variables (n = 303). The occupational category was the only fixed variable that did not meet the proportional hazards assumption. Therefore, the models were stratified on occupational categories, although in the univariate analysis using the Kaplan–Meier curves and log-rank test, occupational categories were related to the probability of RTW (Additional file 1: Fig S2).

In model A, women who underwent breast surgery (HR = 0.08, 95% CI 0.02–0.39), chemotherapy (HR = 0.08, 95% CI 0.03–0.21) and radiotherapy (HR = 0.36, 95% CI 0.19–0.67) were at higher risk of non-RTW. The delivery of antidepressants/anxiolytics was a significant predictor of non-RTW (HR = 0.70, 95%CI 0.51–0.97). In model B, women who received SLC were at higher risk of non-RTW (HR = 0.56, 95% CI 0.31–1.01). In both models, high household income was a RTW facilitator, whereas age > 55 years was a RTW barrier.

This study presents several strengths explained by the quality of the data collected, particularly their temporal and sequential features, using two complementary statistical methods. First, the nationwide CONSTANCES cohort was linked to the SNDS database to obtain exhaustive and dated data on BC treatments, hospital stays, drug deliveries and daily sick pay in the two years after BC diagnosis, thus avoiding self-reported times, memorization bias and attrition. Two different and complementary statistical methods were used to take into account the temporal and sequential features of BC treatments and their related symptoms over time. They gave similar results underlining their robustness. In the literature, the most frequently used models to study RTW are logistic regressions [9, 10, 14, 20, 22, 34] and classical Cox models that do not take into account this temporality and treatment sequences [5, 31]. Instead, a Cox model was chosen in which each BC treatment and drug delivery were treated as time-dependent variables that allowed considering their temporal variations. Furthermore, the sequence analysis method was used to consider the cumulative, order and sequence of different treatments. This method identified five BC care pathway patterns.

Our study presents several limitations. First, we were not able to take into account competitive events, such as retirement, unemployment and disability pension, in our cox models due to the limited number of BCS within each of these situations. Second, reimbursement of anxiolytic/antidepressant and antalgic drugs were used as proxies of depressive symptoms and pain, respectively, in the two years after BC diagnosis. However, these reimbursements could also be related to other physical or mental health issues diagnosed before BC or after BC diagnosis. Third, agricultural and self-employed workers were not included in the CONSTANCES cohort as data on daily sick pay for civil servants are not available in the SNDS. This could have affected the non-RTW rate and the mean duration of sickness absence. However, the percentage of women who did not return to work in the two years after BC diagnosis was similar to the rate reported by a previous French national population-based survey (26.8% versus 25%) [35]. The mean SL duration after BC diagnosis (11.6 months, Additional file 1: Table S1) also was similar to what reported by Fantoni et al. [10] in France and by Balack et al. [5] in the Netherlands (11.5 months). Furthermore, only women with BC and SL > 21 days were included to avoid SL not due to BC. The SL length threshold used in previous studies varied from 14 to 30 days [20, 22]. In our study, women with only short SL (≤ 21 days) received shorter and less complex BC treatments and had less often anxiolytic/antidepressant drug deliveries than women with longer SL (Additional file 1: Table S2). Fourth, the five BC care pathway patterns identified in our study mirror the BC types and treatment protocols, thus minimizing the lack of information on BC stage and targeted therapies [14]. Although hormone therapy and axillary LN dissection are part of BC management, they were not included in the BC care pathways for several reasons. Particularly, most women start hormone therapy when they are already back to work (e.g., 2/3 of hormone therapies were prescribed on year 3 after BC diagnosis in our study) and axillary LN dissections were mainly performed during breast surgery (data not shown).

Finally, BC care pathways were identified only based on BC treatment sequences over time without considering the importance of care and resource organization and coordination [36, 37]. The role of occupational physicians [38], general practitioners [39], breast cancer specialists [40] and of supportive care (i.e., physiotherapists, psychologists, dieticians) after BC diagnosis should also be taken into account to better determine their quantitative and qualitative effect on the RTW process.