The influence on quality of life of intermittent scheduling in first- and second-line chemotherapy of patients with HER2-negative advanced breast cancer

An extensive description of the study design has been published previously [8]. In brief, the Stop&Go study is a phase III trial assessing the impact of scheduling of first- and second-line chemotherapy in patients with HER2-negative advanced breast cancer. Participants were randomized to two sets of four cycles (intermittent) or eight consecutive cycles (continuous) of chemotherapy in first- and second-line treatment. First-line treatment consisted of paclitaxel plus bevacizumab, second-line treatment of capecitabine or non-pegylated liposomal doxorubicin.

QoL was measured by RAND-36 questionnaires [9], which were distributed by post at baseline and every 12 weeks during study treatment and follow-up until death or withdrawal due to any reason. Information about the social status of patient at randomization was collected retrospectively.

The questionnaire contains 36 multiple choice items divided over 8 subscales and is almost identical to the SF-36 [9, 10]. We first calculated the 8 subscales of the RAND-36 using the Dutch algorithm [11]. The weights of this algorithm to calculate T-scores (mean = 50, SD = 10) were obtained in several large Dutch samples (total n = 6822). As there is no Dutch algorithm for the calculation of physical and mental components of QoL out of the RAND-36, we then applied US weights used for the SF-36 to determine these component scores [12]. The physical and mental component scores are presented as T-scores with a mean of 50 and a standard deviation (SD) of 10 for the normal population, with higher scores representing better QoL. Individual missing items of the RAND-36 were handled using the algorithm presented by Ware [13].

The evaluation of QoL comprised a secondary objective of the Stop &Go study. The primary, prospectively planned, objective was to describe the course of physical and mental QoL for both treatment arms, and to estimate differences in changes from baseline in physical and mental QoL between arms. Secondary objectives, added post hoc, were to assess the influence of age, Body Mass Index (BMI), living alone, Eastern Cooperative Oncology Group Performance Status (ECOG PS) at baseline, disease-free interval (DFI) between initial diagnosis and metastatic diagnosis, site of metastasis at baseline, receiving prior hormonal therapy for advanced disease, number and duration of (re-) hospitalizations, duration of first-line treatment, and receiving first- and/or second-line study treatment on QoL and compare these influences between treatment arms.

QoL analyses included all patients within the Stop&Go study who responded to at least one questionnaire with sufficient valid data, according to the intention-to-treat principle. Baseline differences between study arms, dropouts and participants were analyzed with χ² tests with continuity correction, t tests or Mann–Whitney U tests.

Differences in changes from baseline were estimated longitudinally with multilevel linear regression analyses. These analyses can handle data with unbalanced and incomplete time points, assuming missing at random, efficiently. For this multilevel approach, three levels were included; institute as upper level, the patients as middle level and their repeated measures as lower level. The deviance test was applied to determine the necessity of the institute level [14, 15]. For analyses of the treatment effect only, the fixed parts of the models included time, the logarithm of time, treatment group, and the interactions. For the secondary analyses of the covariate effects, separate models were postulated including these covariates with first- and second order interactions.

In order to prevent the few cases with very long follow-up measures to exert an unduly great influence on the models, we determined the 90th percentile of the maximum follow-up time per case. Analyses were performed within the period until this 90th percentile.

For the primary analyses a Bonferroni correction was applied and p values < 0.025 were considered significant. For the exploratory secondary analyses p values < 0.05 were maintained. Based on a systematic review about minimally important differences in health-related QoL studies, an effect size of 0.5 SD (in this case 5 points on the component scores) was considered clinically meaningful [16]. Additionally, we considered a change of 2 points on the component scores a small effect, 5 points a medium effect and 8 points a large effect. [17].

Post hoc power analyses estimated the minimum detectable effect size. We used the observed correlations between the measurements, an alpha of 0.025 and a power of 0.80. As the number of measurements declined rather fast, for each outcome two power analyses were performed: one based on two measurement points and one on four.

Statistical analyses were performed with IBM-SPSS version 24.

At the time of data cut-off for the current analyses (May 1, 2018) a total of 420 patients were included and randomized in the Stop&Go study, of whom 402 (96%) responded to the QoL baseline questionnaire. Due to too much missing items or questionnaires, the population for analyses included 198 patients in the intermittent, and 200 patients in the continuous arm with at least one valid RAND-36 administration (total n = 398) (Fig. 1). There were no statistical significant differences in any of the patient- or treatment characteristics between the arms except for the DFI, which was longer for the intermittent treatment arm (Table 1). No significant difference was found in proportion of patients who continued on second-line treatment (Table 1, Fig. 1). Additionally, no significant differences were observed between the 22 non-responders compared to the 398 responders in treatment arm, early death, age and BMI (data not shown).

The maximum follow-up time was 68.2 months, with a median of 11.4 months (IQR 5.6–22.2). Median number of questionnaires per patient was 4.3 (IQR 2.1–7.5). Considering that 90% of the patients had a follow-up time within the range of 0.0 to 30.6 months, we limited the analyses to 30.6 months (Fig. 1, Supplementary Fig. S1). Drop-out of QoL assessments should take into account that active participants should administer a questionnaire every 12 weeks and may be overdue for half of that period. Thus at the end of the first year, participants should have responded after 34 weeks, or in case they deceased, should have responded at least 18 weeks before they died. At 12 months 297 women were still alive (75%), and 101 (25%) had deceased. Of the alive patients, 56 (19%) dropped out due to unknown reasons, while 21 of the deceased patients (21%) did not administer a questionnaire in their last 18 weeks of life, indicating reasons for drop-out other than death. At 24 months, 180 women were still alive (45%) and 218 had deceased (55%). A total of 76 alive participants (42%) dropped out, and 68 of the deceased (31%) dropped out at least 18 weeks prior to death. At the end of the study period at 30 months, 145 women were still alive (36%), and 253 had died (64%). Eighty-seven patients alive dropped out (60%) and 86 of the deceased patients (34%) dropped out prior to death.

The deviance test pointed out that the institute level was superfluous (χ₍₁₎² = 2.22, p = 0.136). The estimates of the physical and mental component model for the main treatment effect without covariates at the various time points are presented in Table 2. Mean baseline scores were 38.0 and 38.2 for physical, and 45.0 and 42.4 for mental QoL with intermittent and continuous treatment, respectively. The model itself is presented in Supplementary Table S1.

In the intermittent arm there was an almost linear decline of physical QoL causing a clinically meaningful difference of 5.40 points at 24 months (p < 0.001), while scores in the continuous arm stabilized after a decline of ± 3.4 points at 12 months (Table 2, Fig. 2a). Considering mental QoL (Table 2, Fig. 2b), a lower baseline score was observed in the continuous arm (± 2.5 points, p = 0.021) compared to the intermittent arm. Both arms showed an increase in mental health after start of the study treatment (1.58 points; p = 0.005 and 2.48 points; p < 0.001 at 12 months for intermittent and continuous treatment). Similar to the course of physical QoL, the continuous arm had slightly more stable mental QoL scores than the intermittent arm, although for both arms there were no clinically meaningful changes compared to baseline.

When comparing the arms for both components in changes from baseline, the maximum differences were 2.46 points (p = 0.101) for physical and 1.95 points (p = 0.182) for mental scores, both measured at 30 months and in favor of continuous treatment.

Age and BMI had no significantly differential effect on the course of both physical and mental QoL from baseline for the two treatment arms. Duration of first-line treatment did not affect physical and mental QoL scores over time. Regarding the social status of the patients, there were no significant differences between patients that lived alone and those living together for both physical and mental QoL. Likewise, the duration and number of (re)hospitalizations and the treatment line had no effect on both the physical and mental QoL scores over time. The 23 patients with a baseline ECOG PS score of 2 had 13.1-point lower baseline scores on the physical component (p < 0.001) compared to patients with better performance status. These scores increased slightly until 6.8 months, where after there was a large decrease. No significant differences in physical QoL between the treatment arms were observed within this group compared to patients with better performance status (data not shown). Regarding the mental component score, no significant differences were found at all. Concerning the influence of DFI between initial and metastatic diagnosis, patients within the continuous arm had a 0.04 point lower physical component baseline score for each additional month between the initial and metastatic diagnosis (p = 0.007). No further differences were found. For the mental score no relevant significant differences were observed.