Factors associated with medication adherence trajectory among patients treated with oral anticancer drugs: a multicentre cohort study.

Introduction

In France, 106 systemic oral anti-cancer therapies (SACTs) are currently available through the national health service, 39% of which are targeted therapies [ 1 ]. Although oral SACTs offer clear advantages for patients, they are also associated with various drawbacks and therapeutic challenges. One key benefit of taking oral SACTs is the simplification of the treatment pathway, which grants patients greater autonomy and improves quality of life, in contrast to conventional chemotherapy or immunotherapy, which typically require injectable administration and therefore day hospital care [ 2 , 3 ]. Patients become the main actors in their treatment, gaining greater control and responsibility over their care, which can foster empowerment, enhance engagement, and improve adherence [ 4 , 5 ]. This autonomy allows patients to actively participate in decision-making and tailor their management to their personal circumstances, ultimately contributing to better treatment outcomes [ 6 ]. However, this increased responsibility also creates challenges in treatment management. Optimal adherence to oral SACTs may be challenged by several factors, such as cognitive impairment, swallowing difficulties, and complex medication regimens. These barriers are often associated with advanced age or comorbidities, which are common in oncology populations. In addition, polypharmacy increases the risk of drug-drug interactions, including those involving complementary and alternative medicines (CAM), potentially leading to reduced treatment efficacy or increased toxicity [ 7 ]. In outpatient settings, where patients are responsible for managing their treatment, these factors highlight the need for a dynamic and multifactorial approach to understanding adherence behaviours. Although oral SACTs place greater responsibility on patients to follow oncologist-prescribed regimens precisely, high rates of non-adherence have been reported during long-term treatment [ 8 – 10 ]. Despite the importance of medication adherence, its rates vary between patients with different types of cancer and could be as low as 14% for some anti-cancer treatments [ 11 – 13 ] As a result, patients may experience low survival rates [ 14 ], disease progression, increased risk of hospitalisation, greater demand on healthcare resources [ 15 ], reduced functional ability, and poorer quality of life [ 16 ]. Medication adherence is a complex phenomenon influenced by multiple factors: the pathology, the treatments, individual patient factors, environmental influences, and the healthcare systems [ 17 – 19 ].

In this context, assessing medication adherence throughout the patient care pathway is a crucial step in ensuring optimal treatment efficacy. This evaluation remains a medical challenge, as there is no universally accepted gold standard for measuring adherence.

In light of all these considerations, oral SACTs present many challenges, including the need for close and complex patient follow-up. Several studies have already been conducted to identify the most effective approaches to the implementation, understanding, and follow-up of oral therapies. These studies have highlighted, among other things, the indispensable role of the patient’s care and education pathway with a multidisciplinary team [ 12 , 13 , 20 , 21 ].

Given the complexity and variability of adherence behaviours over time, longitudinal approaches are increasingly used to characterise patient adherence patterns. Among these, latent trajectory modelling is a statistical method used to identify unobserved subgroups of subjects who follow similar adherence trajectories during follow-up [ 22 ]. This approach allows for a dynamic assessment of adherence, capturing temporal fluctuations that single-point evaluations may overlook. While these methods have been applied to chronic diseases, their use is limited in the context of oral SACTs [ 22 ].

The overall aim of this study was to gain a better understanding of medication adherence patterns in patients receiving oral SACTs, using longitudinal and multiple measurement methods. The main objective of the study was to identify latent trajectories of medication adherence among patients initiating an oral SACT and to explore the factors associated with allocation to a given trajectory group. A secondary objective was to examine the correlation between a simple six-item adherence Girerd adherence score and therapeutic drug monitoring (TDM) results, in order to assess the validity of the questionnaire within the oral SACTs care pathway.

Methods

Setting and study design

A prospective observational, multicentre cohort study was conducted at the Pompidou Hospital and Cochin Hospital (Academic hospital group Paris Centre, Assistance Publique – Hôpitaux de Paris).

A convenience sample of consecutive patients attending the oncology day hospital for the initiation of an oral SACT at each hospital was invited to participate in the study. Inclusion criteria were (i) treatment for solid tumours with an oral SACT for which TDM was routinely performed, (ii) age over 18 years, (iii) ability to read or understand French, (iv) provision of non-opposition to participation, and (v) attendance at a baseline visit with a clinical pharmacist.

The recruitment period was from February 2020 to March 2021.

Description of the care pathways of patients treated by oral anticancer drugs

In both hospitals, a multidisciplinary risk management programme was implemented as part of routine care for patients initiating an oral SACT during a day hospitalisation. The programme consisted of a pre-therapeutic evaluation, two early follow-up visits (around day 15 and at 1 month after oral SACT initiation), and a late follow-up visit at approximately 6 months after initiation. The core multidisciplinary team was composed of physicians, nurses, clinical pharmacists, and dietitians. Other healthcare professional interventions, including those by psychologists, social workers, and cardiologists, could be consulted based on specific patient needs. The content of the visits performed by clinical pharmacists included (i) ensuring sufficient understanding of the oral SACT administration modalities and self-management of side effects, (ii) detection of drug-related problems such as drug-drug interactions or interactions with CAM, with appropriate pharmaceutical interventions or recommendations provided to the care team when relevant, and (iii) initial evaluation and follow-up (at day 15, at 1 month, and at 6 months) of medication adherence to oral SACTs. The patient came to the oncology day hospital for an initial evaluation when the study participation was proposed to the patients.

Measurements

Girerd adherence scale

The Girerd adherence scale is a six-item self-administered questionnaire assessing medication adherence [ 23 ]. The tool is designed to evaluate medication adherence in patients with hypertension but has been shown to be reliable across a wide variety of medication-adherence studies [ 24 – 27 ]. A score of 0 indicates good medication adherence, a score of 1 or 2 indicates minor non-adherence, and a score of 3 or higher indicates non-adherence. The questionnaire is provided in the supplementary material (Figure S1 ). Authorisation for the use of the Girerd questionnaire for research purposes was obtained on 25/07/2019, in accordance with the licensing conditions required by its author.

Therapeutic drug monitoring

TDM was performed at pharmacokinetic steady state as a standard of care in the two centres at the three follow-up visits. TDMs were centralised in a single laboratory (Unit of Biology of Medicines – Cochin Hospital). Blood samples were collected immediately before the next drug intake during a routine follow-up visit to the day hospital. Plasma drug concentrations were measured using high-performance liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) [ 28 ]. The accuracy of the LC-MS/MS methods was ensured through participation in the external quality assessment scheme provided by the Group of Clinical Pharmacology in Oncology (Unicancer, Paris, France) and Asqualab (Paris, France).

Covariates

Socio-professional status was classified using the French national classification of occupations ( Professions et Catégories Socioprofessionnelles , PCS, from the National Institute of Statistics and Economic Studies—INSEE, France) [ 29 ], which reflects both job type and associated socioeconomic standing. Due to the limited sample size, the original eight-category variable was recoded into three groups, based on the median income levels associated with each occupational group.

Obesity was defined as a body mass index (BMI) ≥ 30 kg/m 2 , in accordance with the World Health Organization (WHO) classification [ 30 ]. Malnutrition was defined according to the diagnostic criteria for adults proposed by the French National Authority for Health ( Haute Autorité de Santé , HAS) [ 31 ]. The Charlson comorbidity index was calculated using clinical data collected at the baseline visit. This index assigns weighted scores to a range of comorbid conditions (e.g. diabetes, chronic pulmonary disease, dementia), reflecting their relative impact on the risk of 1-year mortality. The total score was then used to quantify the overall comorbidity burden [ 32 ]. Polypharmacy was defined as the regular use of five or more medications, based on the patient’s usual prescriptions at the baseline visit. This threshold is commonly used in clinical and epidemiological studies to identify patients at increased risk of adverse drug reactions and medication-related complications [ 33 ]. Eastern Cooperative Oncology Group (ECOG) performance status was recorded at the baseline visit to assess patients’ functional status. This scale ranges from 0 (fully active) to 5 (dead), with higher scores indicating greater functional impairment [ 34 ].

Follow-up outcomes

Disease progression, death, and dose-limiting drug toxicities were assessed at each follow-up visit by the nurse or the referring oncologist. Disease progression was evaluated using the Response Evaluation Criteria In Solid Tumors (RECIST) 1.1 criteria [ 35 ]. Dose-limiting drug toxicities were evaluated by the nurse using the Common Terminology Criteria for Adverse Events (CTCAE) v4.0 scale [ 36 ] in accordance with the guidance provided in the summary of product characteristics for each drug.

Data collection

Data were collected in an electronic case report form (Cleanweb®) including (i) sociodemographic characteristics of the included patients, (ii) clinical and biological data, (iii) the responses to the Girerd questionnaire at day 15, month 1, and month 6.

Ethical considerations

The study was conducted in accordance with good clinical practice guidelines [ 37 ] and French regulations [ 38 ]. The study protocol, patient-information sheet, and non-opposition consent form were approved by the ethics committee (CERAPHP 22/07/19, IDRCB no. 2018-AO3512-53). All potentially eligible patients were provided with detailed study information. Inclusion occurred only after patients gave their non-opposition to participate, as required by French regulatory standards for this type of observational study.

Statistical analysis

Group-based trajectory models

Group-based trajectory modelling (GBTM) is a semi-parametric mixture model that identifies groups of individuals based on latent subgroups within a population that share similar longitudinal trajectories of a single outcome. GBTM was estimated by an Expectation-Maximisation (EM) algorithm, allowing missing data and unbalanced panels [ 39 ]. The selection of the optimal model was based on the lowest value of the Akaike Information Criterion (AIC). The model specifications required to run the analysis included a three-group structure, a linear polynomial order, and standardisation as the normalisation method. Additional specifications for the best GBTM model selection included a minimum trajectory sample size of 5% of the total sample, a minimum average probability of allocation to a given group of 0.7, and the clinical relevance of the groups. The gbmt package [ 40 ] was used to perform the analysis, using RStudio (version 1.4.1106). The time series consisted of a three-point trajectory, comprising the first two early monitoring points and the late monitoring point around month 6.

Analysis populations

The intention-to-treat (ITT) population included all participants enrolled in the cohort. The per-protocol (PP) population was defined as participants who had completed at least two assessments of medication adherence during follow-up—specifically, two medication adherence scores for the main analysis, and both two medication adherence scores and two TDM results for the complementary analysis. The follow-up completer population was defined as the subset of the PP population who completed the final study visit.

Factors associated with adherence groups

A bias-reduced multinomial regression model was used to determine factors associated with allocation to a given medication adherence group ( brmultinom function, brglm2 package, R [ 41 ]). Bivariate bias-reduced multinomial logistic regression models were used to test the association between baseline variables and adherence groups. To assess the global association between explanatory variables and the outcome, univariate multinomial logistic regression models were fitted, and likelihood ratio tests ( Anova function, type III, from the car package , R [ 42 ]) were applied to compare each model to a null model with intercept only. Quantitative variables were dichotomised based on the median value of the PP ( Y = 1 if the value was equal to or above the median, Y = 0 otherwise). All covariates associated with medication adherence groups with a p -value < 0.1 in bivariate analysis were included. A stepwise algorithm with a backward direction was applied to choose the final model, based on the minimisation of the AIC. If necessary, missing data on explanatory variables were handled using multiple imputation by chained equations for variables where the missing at random assumption could be reasonably justified. The continuous optimal adherence group served as the reference group for multivariate analysis. The goodness of fit of the multinomial model was evaluated using the C-statistic, after dichotomising trajectory groups (group 1 vs. other groups, group 2 vs. other groups, and group 3 vs. other groups), a value > 0.7 indicating a good model and a value of 1 indicating a perfect model [ 43 ].

For baseline characteristics and medication adherence group descriptions, data are expressed as mean (standard deviation, SD) or median (interquartile range, IQR) for quantitative variables, and as absolute frequencies ( n ) and percentages (%). To assess the possibility of informative censoring bias—that is, participants censored due to death, toxicity, or disease progression who might be more likely to belong to a specific adherence trajectory group—a logistic regression was conducted to test whether censoring was associated with the trajectory group.

Sensitivity analysis

In order to test the robustness of our results, a sensitivity analysis was performed for multivariable model selection. Dichotomised variables based on the median were further dichotomised using (i) the first quartile (Q1) and (ii) the third quartile (Q3). If a variable was significantly associated with allocation to a given group by bivariate analysis (based on Q1 or Q3), the variable was included in the initial multivariable model.

Validation of the six-item questionnaire using TDM

As an external validation of the Girerd score in the local context, the correlation between the trajectory of medication adherence scores and the trajectory of TDM results was evaluated over time. The GBTM methodology, previously described, was applied to the longitudinal TDM dataset. The TDM results were categorised as “therapeutic range”, “subtherapeutic range”, or “supratherapeutic range” to account for different molecules with varying reference values and to reflect clinically relevant dose adjustment requirements [ 44 ]. Therapeutic levels indicate a likely good level of adherence and no need for dose adjustment; subtherapeutic levels suggest possible adherence issues and the need for dose escalation, while supratherapeutic levels imply the need for dose reduction, especially in cases of toxicities, without indicating low adherence problems [ 44 ]. A contingency table was built by cross-tabulating allocation to medication adherence score groups and TDM groups. The Cramér’s V coefficient was used to assess the association between these two nominal variables. The correlation was considered strong for Cramér’s V values > 0.15 [ 45 ].

Results of the study were reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [ 46 ]. The checklist is provided in the Supplementary Table S1 .

Results

Clinical and demographic data

Patients were included from February 2020 to March 2021. The sociodemographic data and medication adherence of the 135 included patients are presented in Table 1 . The mean age was 63.8 (SD 13.1) years, and 35.6% (48/135) were women. The vast majority of patients had metastatic disease (93.3%), with 73.1% of the subjects having an ECOG performance status of 2 or higher. The median Charlson comorbidity index was 8 (min–max: 2–19), with 26.6% of patients ( n = 34) having a diagnosis of malnutrition. The main tumour localizations were the kidney ( n = 56, 41.5%), joint tumours ( n = 21, 15.6%), and the prostate ( n = 20, 14.8%).

It was found that 50 patients (37.0%) were using CAMs. The most commonly used types were biologically based CAMs, including herbal medicines ( n = 40, 29.6%), vitamin supplements ( n = 11, 8.1%), and aromatherapy ( n = 5, 3.7%). Fourteen subjects (10.4%) were using more than one CAM at the time of oral SACT initiation.

Follow-up

ITT population

From the initial 135 patients included in the study, the mean follow-up duration was 153.4 (SD 7.0) days. In total, 75 participants (55.6%) in the ITT population completed the follow-up, which included the initial inclusion visit, the two early monitoring visits, and the follow-up visit around month 6. The reasons for discontinuation of follow-up were disease progression ( n = 25), oral anticancer drug discontinuation due to toxicity ( n = 13), death ( n = 11), and other reasons ( n = 5). Figure 1 shows the flowchart of the study population.

PP population

The PP population consisted of 71 subjects (51.6% of the total population) who had completed at least two assessments of the medication adherence using the Girerd score during follow-up. In total, 51 participants (71.8%) of the PP population completed the follow-up. Table S2 shows that certain drugs were underrepresented in the PP population compared with the initial ITT sample, suggesting a lower likelihood of full adherence assessment among these patients. In particular, abiraterone, cabozantinib, and regorafenib accounted for 15.8%, 37.3%, and 33.3% of cases in the PP population, respectively, whereas their proportions were higher in the ITT population.

Medication adherence trajectories

The individual medication trajectories of the 71 participants included in the PP analysis are provided in Figure S2 . The GBTM method classified patients into three latent medication adherence trajectory groups (Fig. 2 ): (i) a group of 7 patients with a progressive improvement in medication adherence during follow-up, (ii) a group of 52 patients with consistently good or fairly good adherence during follow-up, and (iii) a group of 12 patients with a decline in medication adherence during follow-up.

Factors associated with medication adherence trajectories

By bivariate analysis (Table 2 ), it was found that the mean Girerd score was significantly higher in group 1 as compared with group 2, at 1.71 (SD = 1.04) and 0.15 (SD = 0.36), respectively ( p < 0.001). Socio-professional status was also significantly associated with the probability of belonging to a medication adherence trajectory (overall p = 0.017), with higher-income individuals (executives, higher intellectual professions, and intermediate professions) having a lower probability of being in the deteriorating adherence group (OR = 0.18; 95% CI 0.04–0.9; p < 0.05). Obesity, defined as a BMI of 30 kg/m 2 or higher, was also associated with an increased likelihood of belonging to the first group, which was characterised by improved medication adherence over the follow-up period ( p = 0.01). Finally, the use of CAMs tended to be associated with medication adherence trajectories (overall p = 0.096), with a higher proportion of users of herbal medicines, vitamin supplements, or aromatherapy in the group exhibiting deteriorating medication adherence. No association was found between group allocation and follow-up outcomes including disease progression, death, or dose-limiting drug toxicity (Table 2 ). Multivariable analysis demonstrated that belonging to the deteriorating adherence group was associated with a higher probability of having a low-income position (OR = 7.9; 95% CI 1.21–51.3; p = 0.03) (Table 3 ). Accuracy of the model was good, with the C-statistic for binary models ranging from 0.69 to 0.86 (Table S3 ). No evidence for informative censoring bias was found, with a non-significant association between the cause of censoring and group allocation ( p = 0.70). No missing data were observed for the covariates included in the multivariable models; therefore, no imputation was performed.

Sensitivity analysis

By bivariate analysis based on Q1 and Q3 dichotomisation, it was found that belonging to the declining adherence group (group 3) was associated with a higher probability of being aged ≥ 71.5 years (OR = 4.0; 95% CI 1.1–15.1; p < 0.05), with group 2 as the reference. When considering BMI cut-offs, the association was no longer statistically significant for patients above 25 kg/m 2 (overweight) and 35 kg/m 2 (severe obesity). However, the same trend was observed: individuals in the improving adherence group compared with the persistent adherent group tended to have a higher proportion of overweight and severely obese participants. The inclusion of age in the initial multivariable model did not change the final model selection after the backward elimination procedure (see Table S4 ).

Validation of the six-item questionnaire using TDM

Individual trajectories for TDM are provided in Figure S3 . Using the GBTM methodology, we identified three trajectory groups. A subsample of n = 40 participants with at least two medication adherence score assessments and two TDM results was used to estimate the correlation between the two trajectories. A high correlation was found, with Cramér’s V = 0.21 (95% CI 0.00–0.42).

Discussion

The use of GBTM methodology enabled the identification of three distinct groups of oral anticancer drugs adherence groups: a first group of improving adherence, a second group of persistent high adherence, and a third group of declining adherence over time. It was found that subjects from the worsening adherence group were less likely to have a high-income position.

Based on longitudinal data, this study found that 83.1% of participants had either persistent good adherence or improving adherence over time. These results are consistent with the level of medication adherence described in patients treated with anticancer drugs, ranging from 46 to 100% [ 47 ], with a commonly accepted threshold of 80% used to define good adherence [ 47 ]. Nevertheless, comparisons should be interpreted with caution, as most published data are derived from cross-sectional studies.

To our knowledge, our study is the first to analyse the medication adherence trajectory to oral SACTs by the GBTM method based on prospectively collected data. In other chronic diseases, a systematic review summarising 21 studies reporting an analysis of medication adherence trajectories showed that studies generally identified four to six trajectory groups [ 48 ]. Studies included in this systematic review were all based on medical administrative databases, with a longitudinal evaluation of medication adherence using the Medication Possession Ratio (MPR) or the Proportion of Days Covered (PDC), which are indirect indicators of medication adherence and are associated with a risk of overestimation and misclassification [ 49 , 50 ]. The limited number of participants included in this prospective study could have reduced the ability of the GBTM method to discriminate between more precise groups, such as initially non-adherent subjects who later achieved improved adherence. However, the three groups identified were clinically relevant in describing the trajectories of medication adherence. There is no consensus on the “gold standard” method for evaluating medication adherence. To address the limitations inherent in individual adherence assessment techniques, the ISPOR Medication Adherence and Persistence Special Interest Group emphasises the importance of combining multiple methods, both direct (e.g. TDM or electronic pillboxes) and indirect (e.g. questionnaires), to reduce the risk of bias, with an approach tailored to the clinical context and local healthcare systems [ 51 ]. Although using a combined approach enhances the reliability and precision of adherence evaluation, it remains underutilised, with only 21% of studies implementing such strategies. In this prospective cohort study, and in line with recommendations [ 52 , 53 ], two complementary methods were used to assess medication adherence: an indirect method, consisting of a validated six-item self-administered questionnaire to assess medication adherence [ 54 ], and a direct method, consisting of iterative TDM. Self-report questionnaires are widely used because they are convenient and enable adherence-related factors and behaviours to be considered [ 55 ]. The biases of patient-reported outcome measures are well known, including acquiescence bias, perception bias, and the Hawthorne effect [ 56 ]. The determination of drug-derived metabolites or the concentration of the drug itself in the serum constitutes an objective and relevant way of exploring drug exposure. However, this technique is relatively costly and is not exempt from so-called white-coat adherence, which occurs when patients change their behaviour before scheduled tests [ 57 ], as well as from inter-patient pharmacokinetic variations. Each of these methods is associated with a risk of bias, but their combination has been shown to be more effective in detecting poorly adherent patients [ 58 ].

No significant association was found between group allocation and follow-up outcomes including disease progression, death, or dose-limiting drug toxicity, which is likely related to the limited sample size. By multivariable analysis, participants from the worsening adherence group were more likely to have a low-income position. This association has been widely described in previous studies [ 59 , 60 ]. In the present study, socio-professional status was used as a proxy for socioeconomic position, based on occupational categories defined by the French national classification (PCS, INSEE). Socioeconomic disadvantage is a well-recognised determinant of poor adherence, as it can influence access to treatment and health literacy [ 61 ]. Several studies have shown that low-income patients are at greater risk of non-adherence, potentially due to limited access to healthcare services, increased psychosocial stress, and competing priorities. For example, a large Korean study demonstrated that even in a universal healthcare system, income-related disparities persisted in medication adherence and health outcomes, possibly due to unmet healthcare needs and greater comorbidity burden [ 62 ].

In bivariate analysis, patients with poorer adherence trajectories tended to report a higher rate of use of biologically based CAMs, including herbal medicines, vitamin supplements, or aromatherapy. The existing evidence regarding the association between CAM use and adherence remains limited and inconsistent: some studies have reported no significant association [ 63 ] while a similar association has been reported in a study conducted in patients with inflammatory bowel disease (that is, an inverse relation between medication adherence and CAM use) [ 64 ]. These conflicting results may be explained by the dynamic dimension of GBTM classification, since participants classified as poor adherents in cross-sectional studies may exhibit different behaviours from those who improve their adherence during follow-up. During the multidisciplinary consultations, clinical pharmacists did not provide proactive counselling on supportive medicines. Instead, they focused on thoroughly collecting information on all substances consumed by the patient, including CAMs, over-the-counter products, and dietary supplements, given the often poorly documented risk of interactions associated with these products. It is also acknowledged that adherence-related behaviours in cancer patients may differ significantly from those observed in chronic conditions, given the severity of the prognosis and the frequent lack of curative therapeutic options in advanced stages of the disease [ 65 ].

This study had some limitations. First, the amount of missing data for medication adherence outcomes (which included the Girerd questionnaire and TDM results) led to a relatively small PP population. This issue can be partially explained by the recruitment period, which coincided with the first wave of the COVID-19 pandemic, thereby limiting face-to-face visits and study assessments. A small PP population may reduce the statistical power of longitudinal analyses and increase the risk of selection bias, as patients included in the final analysis may represent a more adherent or more engaged subgroup. Second, the number of adherence assessments was limited to three time points over a 6-month period. This schedule was determined by the constraints of real-life practice and was intended to minimise loss to follow-up, particularly in a population of patients with metastatic disease and limited life expectancy. Considering the risk of bias, the possibility of informative censoring bias cannot be entirely ruled out, given the limited statistical power associated with the ITT sample. Finally, stepwise selection was used to reduce the number of covariates in the multivariable model. Although commonly applied, this method has well-recognised limitations [ 66 ]. Alternative penalised approaches for variable selection, such as LASSO regression or ridge regression, may provide greater robustness but were not applied in the present study. The final model was fitted using bias-reduction techniques to improve the stability of coefficient estimates.

However, this study also had several strengths. First, the prospective enrolment and follow-up of participants enabled the collection of a broad panel of explanatory variables, including behavioural data such as self-medication, CAM use and type of CAM used, as well as smoking status. Such information, including CAM use and self-medication, is not available in medical administrative databases or retrospective cohort studies. This was underlined by the good predictive accuracy of the multivariable model. Second, using two complementary methods to assess medication adherence and evaluating the correlation between them strengthened the robustness of the findings. Third, the application of a longitudinal methodology allowed a more dynamic assessment of medication adherence. For instance, patients who would have been classified as poorly adherent in cross-sectional analyses were grouped into favourable adherence trajectories using the GBTM approach, as illustrated by participants 6, 24, and 128 (see Figure S2 ). Finally, the risk of selection bias appeared to be limited, given the high rate of acceptance for participation (only 2.2% of eligible patients refused to participate).

To our knowledge, very few studies have explored the use of latent trajectory modelling methods to classify adherence patterns in patients initiating oral SACTs. A recent abstract presented at ASCO 2024 addressed a similar approach in patients with multiple myeloma. Adherence was monitored using an electronic medication event monitoring device. The authors identified three latent trajectories: a high and slightly linear decreasing pattern (62.9%), a high-to-moderate curvilinear decreasing pattern (27.1%) and a low and curvilinear pattern (10%). Ethnicity was found to be significantly associated with trajectory membership, with individuals self-identifying as non-Hispanic Black or Hispanic “other” being more likely to belong to the low-adherence trajectory group [ 67 ]. In the present study, it was not possible to assess the impact of ethnicity, as French regulatory constraints from The Commission Nationale de L’Informatique et des Libertés (CNIL) prohibit the collection of ethnic data [ 68 ].

Such preliminary results warrant replication using claims data from large medical administrative databases, as was recently done in the study of trajectories of medication adherence to low-dose aspirin in the French population [ 64 ], or to nintedanib in patients with idiopathic pulmonary fibrosis [ 69 ]. This groundwork will inform variable selection and facilitate a better understanding of the limitations inherent in retrospective cohort studies using administrative databases. Capturing medication adherence trajectory groups may represent an important step in designing and implementing targeted interventions and programmes to reinforce medication adherence among patients taking oral anticancer drugs.

Conclusion

GBTM classified three distinct adherence trajectory groups reflecting clinically meaningful patterns, particularly socio-professional status as an important determinant of health disparities. Identifying factors associated with group allocation provides a dynamic understanding of the complex phenomenon of medication adherence and may highlight actionable targets to support the development of tailored health interventions.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary file1 (DOCX 176 KB)

Author contribution

• Conceptualization: GP; ATS; VKS; BS • Methodology: GP; VKS • Formal analysis and investigation: GP; VKS • Writing - original draft preparation: GP; VKS • Writing - review and editing: GP; ATS; FG; PBR; ND; CG; JT; RB; BB; JSG; BS; VKS • Funding acquisition: ATS; VKS • Resources: ATS; VKS • Supervision: VKS

Funding

This study was funded by the French Ministry of Health (Grant “Direction Générale de l’Offre de Soins” for the implementation of clinical pharmacy activities in French hospitals).

Data availability

The data that support the findings are available from the corresponding author upon reasonable request.

Declarations

Ethics approval

The study was conducted in accordance with good clinical practice guidelines for biomedical studies [ 36 ] and French regulations. The study protocol, patient-information note, and non-opposition consent form were approved by an ethics committee (CERAPHP 22/07/19, n°IDRCB 2018-AO3512-53).

Consent to participate

All eligible patients agreed to participate in the study and signed a non-opposition consent form.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher's Note

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

References

1. Développement des anticancéreux oraux - Projections à court, moyen et long termes - Ref : ETATDEVANTIKOR17 [Internet]. [cited 2024 Jun 21]. Available from: https://www.e-cancer.fr/Expertises-et-publications/Catalogue-des-publications/Developpement-des-anticancereux-oraux-Projections-a-court-moyen-et-long-termes

2. Coudert P (2022) Les chimiothérapies orales, une avancée dans le traitement des cancers. Actual Pharm 61(612):23–26. 10.1016/j.actpha.2021.12.006

3. Borner M, Scheithauer W, Twelves C, Maroun J, Wilke H (2001) Answering patients’ needs: oral alternatives to intravenous therapy. Oncologist 6(S4):12–16. 10.1634/theoncologist.6-suppl_4-12

4. Aubert I, Kletz F, Sardas J-C (2023) The patient as an actor in his care pathway: insights from the French case. Health Serv Insights 16:11786329231196028. 10.1177/11786329231196029

5. Naik AD, Dyer CB, Kunik ME, McCullough LB (2009) Patient autonomy for the management of chronic conditions: a two-component re-conceptualization. Am J Bioeth 9:23–30. 10.1080/15265160802654111

6. Mason M, Harris MR, Greer JA, Jiang Y (2022) A concept analysis of oral anticancer agent self-management. Cancer Nurs 45:E374–E387. 10.1097/NCC.0000000000000934

7. Werneke U, Earl J, Seydel C, Horn O, Crichton P, Fannon D (2004) Potential health risks of complementary alternative medicines in cancer patients. Br J Cancer 90:408–413. 10.1038/sj.bjc.6601560

8. Yoshida C, Komeno T, Hori M, Kimura T, Fujii M, Okoshi Y et al (2011) Adherence to the standard dose of imatinib, rather than dose adjustment based on its plasma concentration, is critical to achieve a deep molecular response in patients with chronic myeloid leukemia. Int J Hematol 93:618–623. 10.1007/s12185-011-0838-3

9. Verbrugghe M, Verhaeghe S, Lauwaert K, Beeckman D, Van Hecke A (2013) Determinants and associated factors influencing medication adherence and persistence to oral anticancer drugs: a systematic review. Cancer Treat Rev 39:610–621. 10.1016/j.ctrv.2012.12.014

10. Geynisman DM, Wickersham KE (2013) Adherence to targeted oral anticancer medications. Discov Med. 15:231–41 ( PMID: 23636140; PMCID: PMC6477693 )

11. Partridge AH, Wang PS, Winer EP, Avorn J (2003) Nonadherence to adjuvant tamoxifen therapy in women with primary breast cancer. J Clin Oncol 21:602–606. 10.1200/JCO.2003.07.071

12. Felton MA, Van Londen G, Marcum ZA (2016) Medication adherence to oral cancer therapy: the promising role of the pharmacist. J Oncol Pharm Pract 22:378–381. 10.1177/1078155214559114

13. Joy AM, Up N, Chand S, Shetty KJ, George SM, Chacko CS et al (2021) Role of clinical pharmacist in the medication adherence behaviour of cancer patients: An interventional study. Pharm Hosp Clin 56:291–297. 10.1016/j.phclin.2021.05.002

14. Hershman DL, Shao T, Kushi LH, Buono D, Tsai WY, Fehrenbacher L et al (2011) Early discontinuation and non-adherence to adjuvant hormonal therapy are associated with increased mortality in women with breast cancer. Breast Cancer Res Treat 126:529–537. 10.1007/s10549-010-1132-4

15. Kengne AP, Brière J-B, Zhu L, Li J, Bhatia MK, Atanasov P et al (2024) Impact of poor medication adherence on clinical outcomes and health resource utilization in patients with hypertension and/or dyslipidemia: systematic review. Expert Rev Pharmacoecon Outcomes Res 24:143–154. 10.1080/14737167.2023.2266135

16. Ganesan P, Sagar TG, Dubashi B, Rajendranath R, Kannan K, Cyriac S et al (2011) Nonadherence to Imatinib adversely affects event free survival in chronic phase chronic myeloid leukemia. Am J Hematol 86:471–474. 10.1002/ajh.22019

17. Dang TH, Forkan ARM, Wickramasinghe N, Jayaraman PP, Alexander M, Burbury K et al (2022) Investigation of intervention solutions to enhance adherence to oral anticancer medicines in adults: overview of reviews. JMIR Cancer 8:e34833. 10.2196/34833

18. Kavookjian J, Wittayanukorn S (2015) Interventions for adherence with oral chemotherapy in hematological malignancies: a systematic review. Res Social Adm Pharm 11:303–314. 10.1016/j.sapharm.2014.08.006

19. Organization WH. Adherence to long-term therapies : evidence for action [Internet]. World Health Organization; 2003 [cited 2025 Jun 20]. Available from: https://iris.who.int/handle/10665/42682

20. Colombo LRP, Aguiar PM, Lima TM, Storpirtis S (2017) The effects of pharmacist interventions on adult outpatients with cancer: a systematic review. J Clin Pharm Ther 42(4):414–424. 10.1111/jcpt.12562

21. Mawhinney MW, Warden J, Stoner N (2019) The oral education clinic: a pharmacist- and nurse-led clinic to support patients starting oral systemic anti-cancer treatments. J Oncol Pharm Pract 25:449–453. 10.1177/1078155217727820

22. Nguena Nguefack HL, Pagé MG, Katz J, Choinière M, Vanasse A, Dorais M et al (2020) Trajectory modelling techniques useful to epidemiological research: a comparative narrative review of approaches. Clin Epidemiol Volume 12:1205–1222. 10.2147/CLEP.S265287

23. Girerd X, Hanon O, Anagnostopoulos K, Ciupek C, Mourad JJ, Consoli S (2001) Assessment of antihypertensive compliance using a self-administered questionnaire: development and use in a hypertension clinic. Presse Med Paris Fr 30:1044–1048

24. Sapoval M, Thiounn N, Descazeaud A, Déan C, Ruffion A, Pagnoux G et al (2023) Prostatic artery embolisation versus medical treatment in patients with benign prostatic hyperplasia (PARTEM): a randomised, multicentre, open-label, phase 3, superiority trial. The Lancet Regional Health 31:100672. 10.1016/j.lanepe.2023.100672

25. Sustersic M, Gauchet A, Duvert A, Gonnet L, Foote A, Vermorel C et al (2019) Proposal for a global adherence scale for acute conditions (GASAC): a prospective cohort study in two emergency departments. PLoS ONE 14:e0215415. 10.1371/journal.pone.0215415

26. Tardy C, Puszkiel A, Boudou-Rouquette P, De Percin S, Alexandre J, Berge M et al (2024) Pazopanib pharmacokinetically guided dose optimization in three cancer patients with gastrointestinal resection. Cancer Chemother Pharmacol 93:169–175. 10.1007/s00280-023-04574-z

27. Monchablon C, Gondé H, Pouplin S, Varin R, Vittecoq O, Lequerré T (2020) Assessment of adherence to disease-modifying anti-rheumatic drugs in rheumatoid arthritis. Clin Rheumatol 39:207–216. 10.1007/s10067-019-04837-2

28. Balakirouchenane D, Khoudour N, Guégan S, Kramkimel N, Franck N, Rodier T et al (2021) Simultaneous quantification of dabrafenib, hydroxy-dabrafenib and trametinib in human plasma by liquid chromatography-tandem mass spectrometry. J Pharm Biomed Anal 193:113718. 10.1016/j.jpba.2020.113718

29. Nomenclatures des professions et catégories socioprofessionnelles | Insee [Internet]. [cited 2025 Jun 20]. Available from: https://www.insee.fr/fr/information/2406153

30. Organization WH. WHO European regional obesity report 2022. WHO Eur Reg Obes Rep 2022 [Internet]. 2022 [cited 2025 Aug 5]. Available from: https://pesquisa.bvsalud.org/portal/resource/pt/who-353747

31. Diagnostic de la dénutrition de l’enfant et de l’adulte [Internet]. Haute Aut. Santé. [cited 2025 Aug 5]. Available from: https://www.has-sante.fr/jcms/p_3118872/fr/diagnostic-de-la-denutrition-de-l-enfant-et-de-l-adulte

32. Charlson ME, Carrozzino D, Guidi J, Patierno C (2022) Charlson comorbidity index: a critical review of clinimetric properties. Psychother Psychosom 91:8–35. 10.1159/000521288

33. Masnoon N, Shakib S, Kalisch-Ellett L, Caughey GE (2017) What is polypharmacy? A systematic review of definitions. BMC Geriatr 17:230. 10.1186/s12877-017-0621-2

34. Oken MM, Creech RH, Tormey DC, Horton J, Davis TE, McFadden ET et al (1982) Toxicity and response criteria of the Eastern Cooperative Oncology Group. Am J Clin Oncol 5:649–655

35. Schwartz LH, Litière S, de Vries E, Ford R, Gwyther S, Mandrekar S et al (2016) RECIST 1.1-update and clarification: from the RECIST committee. Eur J Cancer 62:132–137. 10.1016/j.ejca.2016.03.081

36. Common Terminology Criteria for Adverse Events (CTCAE) | Protocol Development | CTEP [Internet]. [cited 2022 May 6]. Available from: https://ctep.cancer.gov/protocoldevelopment/electronic_applications/ctc.htm

37. Décret n° 2016–1537 du 16 Novembre 2016 relatif aux recherches impliquant la personne humaine. 2016–1537 Nov 16, 2016.

38. Nagin DS, Jones BL, Passos VL, Tremblay RE (2018) Group-based multi-trajectory modeling. Stat Methods Med Res 27:2015–2023. 10.1177/0962280216673085

39. Alessandro Magrini. gbmt: group-based multivariate trajectory modeling [Internet]. 2021 [cited 2024 Jun 21]. p. 0.1.3. 10.32614/CRAN.package.gbmtAvailable from: https://CRAN.R-project.org/package=gbmt

40. Ripley B, Venables W. nnet: feed-forward neural networks and multinomial log-linear models [Internet]. 2025 [cited 2025 Jun 20]. Available from: https://cran.r-project.org/web/packages/nnet/index.html

41. Fox J, Weisberg S, Price B, Adler D, Bates D, Baud-Bovy G, et al. car: Companion to applied regression [Internet]. 2024 [cited 2025 Aug 5]. Available from: https://cran.r-project.org/web/packages/car/index.html

42. Pencina MJ, D’Agostino RB (2015) Evaluating discrimination of risk prediction models: the C statistic. JAMA 314:1063. 10.1001/jama.2015.11082

43. Mueller-Schoell A, Groenland SL, Scherf-Clavel O, van Dyk M, Huisinga W, Michelet R et al (2021) Therapeutic drug monitoring of oral targeted antineoplastic drugs. Eur J Clin Pharmacol 77:441–464. 10.1007/s00228-020-03014-8

44. Akoglu H (2018) User’s guide to correlation coefficients. Turk J Emerg Med 18:91–93. 10.1016/j.tjem.2018.08.001

45. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP et al (2008) The strengthening the reporting of observational studies in epidemiology (STROBE) statement: guidelines for reporting observational studies. J Clin Epidemiol 61:344–349. 10.1016/j.jclinepi.2007.11.008

46. Karve S, Cleves MA, Helm M, Hudson TJ, West DS, Martin BC (2009) Good and poor adherence: optimal cut-point for adherence measures using administrative claims data. Curr Med Res Opin 25:2303–2310. 10.1185/03007990903126833

47. Alhazami M, Pontinha VM, Patterson JA, Holdford DA (2020) Medication adherence trajectories: a systematic literature review. J Manag Care Spec Pharm 26:1138–1152. 10.18553/jmcp.2020.26.9.1138

48. Martin BC, Wiley-Exley EK, Richards S, Domino ME, Carey TS, Sleath BL (2009) Contrasting measures of adherence with simple drug use, medication switching, and therapeutic duplication. Ann Pharmacother 43:36–44. 10.1345/aph.1K671

49. Vrijens B, De Geest S, Hughes DA, Przemyslaw K, Demonceau J, Ruppar T et al (2012) A new taxonomy for describing and defining adherence to medications. Br J Clin Pharmacol 73:691–705. 10.1111/j.1365-2125.2012.04167.x

50. Differences between MPR and PDC | OpenHealth Knowledge Base [Internet]. [cited 2024 Jun 21]. Available from: https://success.openhealth.fr/en/articles/3722666-differences-between-mpr-and-pdc

51. Ágh T, Hiligsmann M, Borah B, Beaudart C, Turcu-Stiolica A, Manias E et al (2024) Systematic review of outcomes for assessment of medication adherence enhancing interventions: an ISPOR special interest group report. Value Health 27:133–142. 10.1016/j.jval.2023.10.016

52. Allenet B, Baudrant M, Lehmann A, Gauchet A, Roustit M, Bedouch P et al (2013) Comment évaluer l’adhésion médicamenteuse ? Le point sur les méthodes. Ann Pharm Fr 71:135–141. 10.1016/j.pharma.2012.10.001

53. Effoe VS, Suerken CK, Quandt SA, Bell RA, Arcury TA (2017) The association of complementary therapy use with prescription medication adherence among older community-dwelling adults. J Appl Gerontol 36:1054–1069. 10.1177/0733464815602116

54. Tegegn HG, Wark S, Tursan d’Espaignet E, Spark MJ (2022) Measurement properties of patient-reported outcome measures for medication adherence in cardiovascular disease: a COSMIN systematic review. Clin Drug Investig 42:879–908. 10.1007/s40261-022-01199-7

55. Zini MLL, Banfi G (2021) A narrative literature review of bias in collecting patient reported outcomes measures (PROMs). Int J Environ Res Public Health 18:12445. 10.3390/ijerph182312445

56. Podsadecki TJ, Vrijens BC, Tousset EP, Rode RA, Hanna GJ (2008) “White coat compliance” limits the reliability of therapeutic drug monitoring in HIV-1-infected patients. HIV Clin Trials 9:238–246. 10.1310/hct0904-238

57. Wouters H, Rhebergen D, Vervloet M, Egberts A, Taxis K, Van Dijk L et al (2019) Distinct profiles on subjective and objective adherence measures in patients prescribed antidepressants. Drugs 79:647–654. 10.1007/s40265-019-01107-y

58. Salam T, Desai U, Lefebvre P, Jian-Yu E, Greatsinger A, Zacharia N et al (2023) Unintended consequences of increased out-of-pocket costs during Medicare coverage gap on anticoagulant discontinuation and stroke. Adv Ther 40:4523–4544. 10.1007/s12325-023-02620-z

59. Polinski JM, Shrank WH, Huskamp HA, Glynn RJ, Liberman JN, Schneeweiss S, Stuckler D, editor (2011) Changes in drug utilization during a gap in insurance coverage: an examination of the Medicare part D coverage gap. PLoS Med 8:e1001075. 10.1371/journal.pmed.1001075

60. Coughlin SS, Vernon M, Hatzigeorgiou C, George V (2020) Health literacy, social determinants of health, and disease prevention and control. J Environ Health Sci. 6:3061 ( PMID: 33604453; PMCID: PMC7889072 )

61. Lee H, Park JH, Floyd JS, Park S, Kim HC (2019) Combined effect of income and medication adherence on mortality in newly treated hypertension: nationwide study of 16 million person-years. J Am Heart Assoc 8:e013148. 10.1161/JAHA.119.013148

62. Almakadma AH, De Vol A, Alabdaljabar MS, Aldosari S, Muhsen I, AlFreihi O et al (2023) Complementary and alternative medicine use and its association with medication adherence in inflammatory bowel disease and other gastrointestinal diseases. Saudi J Gastroenterol 29:233–239. 10.4103/sjg.sjg_468_22

63. Ajrouche A, Estellat C, De Rycke Y, Tubach F (2020) Trajectories of adherence to low-dose aspirin treatment among the French population. J Cardiovasc Pharmacol Ther 25:37–46. 10.1177/1074248419865287

64. Hlubocky FJ, Ratain MJ, Wen M, Daugherty CK (2007) Complementary and alternative medicine among advanced cancer patients enrolled on phase I trials: a study of prognosis, quality of life, and preferences for decision making. J Clin Oncol 25(5):548–554. 10.1200/JCO.2005.03.9800

65. Smith G (2018) Step away from stepwise. J Big Data 5:32. 10.1186/s40537-018-0143-6

66. Belcher SM, Sereika SM, Dunbar-Jacob J, Yeager K, Rosenzweig MQ, Agha ME et al (2024) Patterns and predictors of electronically measured oral anticancer medication (OAM) adherence among patients with multiple myeloma (MM). J Clin Oncol 42:7547–7547. 10.1200/JCO.2024.42.16_suppl.7547

67. Commission nationale de l’informatique et des libertés (CNIL). Décision 2015–061 du 24 juin 2015 [Internet]. 2015‑061 24, 2015. Available from: https://www.legifrance.gouv.fr/cnil/id/CNILTEXT000030950817 . [cited 2025 Sep 1]

68. Nili M, Epstein AJ, Nunag D, Olson A, Borah B (2023) Using group based trajectory modeling for assessing medication adherence to nintedanib among idiopathic pulmonary fibrosis patients. BMC Pulm Med 23:230. 10.1186/s12890-023-02496-3

69. ICH E6 Good clinical practice - Scientific guideline | European Medicines Agency (EMA) [Internet]. 2002 [cited 2025 Aug 5]. Available from: https://www.ema.europa.eu/en/ich-e6-good-clinical-practice-scientific-guideline

Associated Data

Supplementary Materials

Supplementary file1 (DOCX 176 KB)

Data Availability Statement

The data that support the findings are available from the corresponding author upon reasonable request.