This was an observational, prospective, multicenter study performed by 6 hospital pharmacists in Spain. This study was approved by the Ethics Committee for Research with Medicines from Galicia (Santiago de Compostela, Spain), and all patients provided written informed consent.

Patients were included if they were 18 years or older, had been diagnosed with locally advanced or metastatic NSCLC, and were receiving or prescribed OAT for the management of NSCLC by the oncologist. Patients were excluded if, according to the investigator’s judgment, they were unable to understand the questionnaire administered during the study.

After inclusion, patients were actively followed-up over three months with four study visits, at baseline, and every month. Twelve months after patient inclusion, information on disease progression and OAT prescriptions and dispensations were recorded. At the time of inclusion, the pharmacist recorded information on demographics; disease-related variables, such as date of diagnosis; disease stage at the time of diagnosis; histology; ECOG performance status; number and location of metastases; biomarkers (specifically ALK, EGFR, EGFR-T790M, ROS-1, and BRAF); and treatment-related variables, including previous antineoplastic agents, concomitant medication, and OATs. At monthly visits, we recorded information on pill count, pharmaceutical interventions, and resource utilization, and patients completed the 4-item Morisky Green Levine Medication Adherence Scale (MGLS-4) (Jiménez et al. 1992). In addition, at the follow-up visits, patients completed other patient-reported outcomes (PROs), including the Brief Illness Perception Questionnaire (BIPQ) (Pacheco-Huergo et al. 2012) and the 3-level version of the EQ-5D (EQ-5D-3L) (Badia et al. 1999), which were administered at baseline and month 3; the results of these and other PROs will be published elsewhere.

The MGLS-4 is a brief self-administered questionnaire widely used to evaluate adherence in patients with chronic health conditions; it includes 4 yes/no questions on treatment adherence related to forgetfulness or the disease course (Jiménez et al. 1992). Each question is rated as 0 (no) or 1 (yes), and a score equal to or greater than 1 implies some degree of nonadherence.

The sample size was based on the estimation of a proportion using the maximum variability criterion (i.e., an estimated prevalence of 50%) and a precision of 10%. Assuming a 10% loss, the sample size was estimated to be 106.

The primary outcome was the proportion of treatment adherence as evaluated by pill reconciliation during the 3-month active follow-up period. This parameter was calculated as the difference between the number of pills dispensed and the number of pills returned in the numerator, divided by the product of the number of days of treatment by the number of pills/day prescribed by the oncologist in the denominator, and multiplied by 100 to obtain the percentage of adherence. Patients with a percentage adherence > 80% were categorized as adherent and are referred to hereafter as such. Adherence based on a cutoff value of 90% was also calculated.

Adherence was also estimated via the proportion of days covered (PDC) and the MGLS-4. The PDC was calculated as the sum of days covered by the drug based on the prescription fill date and days of supply divided by the number of days of the treatment period (counted as the number of days between the index prescription dates until the end of the year of the study, study discontinuation, or death) and was multiplied by 100. Patients with a PDC ≥ 80% were considered adherent. Using the MGLS-4, patients were considered adherent if their score was 0.

Adherence as a continuous outcome is presented as the mean and standard deviation and as a binary outcome with absolute and relative frequencies and the corresponding 95% confidence interval.

We analyzed adherence to the following prespecified treatment subgroups: first- and second-generation EGFR-TKIs (erlotinib, afatinib and gefitinib), third-generation EGFR-TKIs (osimertinib), and ALK inhibitors (alectinib, crizotinib and lorlatinib).

The factors potentially associated with treatment adherence were explored via multiple logistic regression analysis, where the dependent variable was treatment adherence, as defined for the primary outcome, and the independent variables were as follows: age (continuous), sex, the presence of a person who reminds the patient daily or weekly to take the medication (binary), the anxiety‒depression item of the EQ-5D-3L (categorical), the ‘personal control’ item (“How much control do you feel you have over your illness?) of the BIPQ (continuous), the ‘treatment control’ item (‘How much do you think your treatment can help your illness?’) of the BIPQ (continuous), the ‘understanding’ item (‘How well do you feel you understand your illness?’) of the BIPQ (continuous), the number of concomitant medications (continuous), and the time since the first administration of oral antineoplastic treatment (continuous). This model was repeated excluding the time since the first administration of the oral antineoplastic treatment (months).

To explore the potential relationship between treatment adherence and progression-free survival, we used the Kaplan‒Meier method and the log-rank test to compare adherent and nonadherent patients based on pill count, and the effect size was calculated via univariate Cox regression analysis. Survival analysis was performed from the time of patient inclusion until 12 months after inclusion; as a sensitivity analysis, we also analyzed the time to progression from treatment initiation. In addition, we performed a post hoc multivariate Cox regression analysis to explore the factors associated with progression-free survival with the following independent variables: age (≤ 65/ > 65 years), sex (binary), Eastern Cooperative Oncology Group (ECOG-PS) score (categorical), tobacco consumption (categorical), presence of locally advanced/metastatic disease (binary), presence of central nervous system metastases (binary), molecular classification (binary: EGFR/ALK) and treatment adherence, as defined in the primary outcome (binary). As a sensitivity analysis, we also performed multivariate Cox regression analysis considering the time to progression from treatment initiation. A stepwise backward approach was used for fitting the regression models. All analyses were performed via IBM SPSS Statistics 26.0 and were considered significant if the p value was < 0.05.