Using population-based data to evaluate the impact of adherence to endocrine therapy on survival in breast cancer through the web-application BreCanSurvPred

Abstract

We show how the use and interpretation of population-based cancer survival indicators can help oncologists talk with breast cancer (BC) patients about the relationship between their prognosis and their adherence to endocrine therapy (ET). The study population comprised a population-based cohort of estrogen receptor positive BC patients (N = 1268) diagnosed in Girona and Tarragona (Northeastern Spain) and classified according to HER2 status (+ / -), stage at diagnosis (I/II/III) and five-year cumulative adherence rate (adherent > 80%; non-adherent ≤ 80%). Cox regression analysis was performed to identify significant prognostic factors for overall survival, whereas relative survival (RS) was used to estimate the crude probability of death due to BC (P_BC). Stage and adherence to ET were the significant factors for predicting all-cause mortality. Compared to stage I, risk of death increased in stage II (hazard ratio [HR] 2.24, 95% confidence interval [CI]: 1.51-3.30) and stage III (HR 5.11, 95% CI 3.46-7.51), and it decreased with adherence to ET (HR 0.57, 95% CI 0.41-0.59). P_BC differences were higher in non-adherent patients compared to adherent ones and increased across stages: stage I: 6.61% (95% CI 0.05-13.20); stage II: 9.77% (95% CI 0.59-19.01), and stage III: 22.31% (95% CI 6.34-38.45). The age-adjusted survival curves derived from this modeling were implemented in the web application BreCanSurvPred ( https://pdocomputation.snpstats.net/BreCanSurvPred ). Web applications like BreCanSurvPred can help oncologists discuss the consequences of non-adherence to prescribed ET with patients.

Figure 1

Patients diagnosed with hormone receptor-positive breast cancer before the age of 85 years in Girona and Tarragona: age-adjusted survival curves according to (a) age groups, (b) HER2 status, (c) stage at diagnosis, and (d) adherence to endocrine treatment.

Figure 2

Plot of the time-dependent effect of “adherence” with 95% confidence intervals: in blue, log-hazard ratio coefficient, β_adherence, estimated from the model with splines (model C.3) considering adherence as constant effect; in red, “time-varying” effect of “adherence” estimated through LOWESS regression of the scaled Schoenfeld residuals from model C.3 versus time (follow-up).

Figure 3

Predicted cumulative probabilities of death due to cancer at 5 (panels a–c) and 10 years (panels d–f) after diagnosis according to age and adherence to treatment (results presented by stage of BC at diagnosis).

Figure 4

Snapshot of the web-based survival prediction application BreCanSurvPred. This snapshot demonstrates the probabilities of survival and death as well as the 5-year conditional probabilities of observed survival and relative survival for a 60-year old patient who was not adherent to endocrine therapy and who was diagnosed with stage III molecular subtype HER2 − /HR + breast cancer. These probabilites are calculated up to 10 years after BC diagnosis.

Figure 5

Snapshot of the web-based survival prediction application BreCanSurvPred. This snapshot shows the comparison of survival according to adherence to endocrine treatment for a 60-year old patient diagnosed with molecular subtype HER2 − /HR + in stage III. These survival probabilites are calculated up to 10 years after BC diagnosis. (Time: time since diagnosis [follow-up]; Percentage: percent survival at a certain Time ).