The study protocol was approved by CPRD’s Independent Scientific Advisory Committee (ID: 21_000411). In the context of CPRD data usage, informed consent is not required for the anonymized data provided to CPRD from participating GP practices. The data is anonymized to the extent that individual patients cannot be identified. Participants have the option to opt out of having their health records shared for research purposes [19]. The original study approval covers secondary analyses, eliminating the need for additional consent from participants. No compensation was provided to participants in this study. Approved researchers could gain access to the data after submitting a study protocol.

This was an observational study using longitudinal data on prescribed medication (statins) extracted from electronic health record data to compare patient-level aggregate adherence 12 months before and 12 months after patient portal registration, Adherence was measured using the medication possession ratio (MPR) with a threshold of 80% as an indicator of adherence. The data source, CPRD Aurum, comprises electronic health record data from GPs in England, where GPs are the gatekeepers of the health care system, and almost all of the population is registered with a GP [20]. The study period varied by patient and ranged from January 1, 2014, to August 15, 2021.

To determine our population, we first included all patients in CPRD Aurum who had a patient portal registration code (available in Multimedia Appendix 1) recorded on or after January 1, 2015, when online services were required to be offered by all GPs in England [6]. Eligible patients had a diagnosis of at least one of the following conditions: adverse cardiovascular event, CKD, CVD, or diabetes, using codes in the study by Davidson et al [21]. These patient groups were chosen because they may be prescribed statins according to National Institute for Health and Care Excellence (NICE) guidelines for CVD prevention [22]. Although not all patients with diabetes will be prescribed statins, we included them due to the interest in the literature on the association between patient portal use and diabetes-related outcomes, including statin adherence [23].

All analyses were performed using RStudio software (version 2023.06.0+421; R Foundation for Statistical Computing). We compared statin prescription refill adherence (measured by MPR ≥80%) 12 months before and 12 months after patient portal registration using multilevel logistic regression. We assumed that once patients were prescribed statins they would be for long-term use, indicating that they should have ordered statins consistently over the 12-month period before and after patient portal registration. The multilevel logistic regression was performed in three major steps:

We reduced the follow up time to 6 months instead of 12 months before and after patient portal registration to examine potential differences in association. Additionally, we performed a sensitivity analysis including all patients that were initially excluded from the study due to missing IMD measure data.

Patients who had missing duration or 0-duration prescriptions were excluded (n=6180) (Figure 1). We excluded all patients whose first statin prescription was recorded at least 12 months before patient portal registration date (n=22,966) (Figure 1) to ensure that patients had at least 12 months of follow-up before patient portal registration. All patients whose statin observations did not fall within the required follow-up period (12 months before and 12 months after patient portal registration date) were excluded (n=7595) (Figure 1). Finally, we only included patients with complete data in the analysis, and excluded all patients that had missing IMD quintile data (n=44,141) (Figure 1). There were no missing data in any other variable except ethnicity, which was included as a separate category in the analysis.

Figure 1 is reused from PhD Thesis by Alturkistani [18].

A total of 44,141 patients were included in the study. The majority (n=13,812) of the patients were men aged 65‐74 years at the time of patient portal registration and had a prescription refill adherence of 80% or higher (n=27,208) (Table 2).

The mean MPR 12 months before patient portal registration was 87% and ranged from 0%-183%, while the mean MPR 12 months after patient portal registration was 81%, ranging from 0%-164%. The median MPR was 92%, both before and after patient portal registration.

Table 2 presents the results of the fully adjusted, multilevel logistic regression for the outcome variable (statin prescription refill adherence of ≥50%, ≥65%, and ≥80%). The odds ratio (OR) for statin prescription refill adherence of ≥80% was 0.84 (95% CI 0.81-0.86), indicating a 16% reduction in the odds of statin prescription refill adherence at this threshold after patient portal registration, compared to 12 months before registration (Table 3). The OR for adherence at 50% and 65% thresholds also indicated a reduction in adherence after patient portal registration, indicating that even patients with lower baseline adherence were less likely to remain adherent after patient portal registration (Table 3). The ICC varied for different levels of adherence but were generally 15% or lower, indicating the proportion of adherence variance that can be attributable to practice-level factors. For example, an ICC of 15% can be interpreted as 15% of the total variance in adherence by patients can be explained by differences at the GP practice level.

When the time of follow up was reduced to 6 months, the OR for statin prescription refill adherence of ≥80% was 1.01 (95% CI 0.98-1.04) in the fully-adjusted model, indicating no difference in statin prescription refill adherence 6 months after patient portal registration compared to the 6 months before patient portal registration (Multimedia Appendix 2).

The OR differed by only 0.01‐0.02 of sensitivity analyses, which included patients who had missing IMD quintiles data (results not shown).

After registering to the patient portal, patients had lower odds of achieving adherence of 50%, 65%, and 80% thresholds. However, this relationship was attenuated when the follow-up time was reduced to 6 months before and after patient portal registration and indicated no change in adherence after registration. This study found a reduction in prescription refill adherence after patient portal registration, which may be due to several reasons. Patients in this study were only registered to the portal once, and it is likely that may have enrolled at a time of high health care utilization. This could lead to increased medication requests prior to patient portal registration, compared to after registration. Additionally, patients may have overall reduced adherence to statins after using it for some time due to factors such as side effects, continuity of care, disease severity, or socioeconomic factors, which could all affect adherence to ordering prescriptions [26,27].

Previous studies revealed that factors such as experiencing side effects, CVD severity, continuity of care and income, or other indicators of deprivation may influence adherence to statins [26,27]. Other factors that could be associated with statin adherence but were not examined in this study include the type of statin, intensity of the dose, timing of prescribing statins (eg, before or after puberty) [17]. Given that statin adherence reduced when patients were followed up for one year but not over 6 months, it is possible that statin adherence naturally reduced over time. Since the codes used for patient portal registration in this study indicated to access medical records, we could not explore potential mechanisms involved between accessing medical records and medication adherence. We acknowledge this as a limitation of the study, as registration to the patient portal does not necessarily mean continued use of the patient portal, discussed further in the limitations and strengths subsection [35].

Among features of patient portals, the ability to order prescriptions through the portal is most strongly associated with medication adherence. For example, one study reported that when medication is ordered through a patient portal, it can have less errors and more clear instructions (compared to hand-written prescriptions), which in turn improves adherence to treatment [36]. Having a higher supply of statin (eg, a supply of 60 d vs 30 d) was associated with improved adherence, which can be achieved by using a patient portal (by providing instant access to ordering medication) [17].

This study is relevant in the wake of increased popularity of using patient portals and other remote or digital health care services, particularly during and post-COVID-19 pandemic, both in England and other countries. Although widely offered within the NHS, this technology is rarely studied. We were unable to study active or continuous patient portal use, as this information was not recorded in CPRD, despite it being a rich source of data for clinical patient records. This study highlights the need to improve recording of the use of technologies to enable research in this field. To the authors’ knowledge, this is the first study using CPRD data to examine patient portal use. Although the study was only able to capture registration to patient portals, the records of registration were obtained from an objective source (ie, electronic health records), which is a reliable method.

A strength of this study was that it controlled for clustering of patients within GP practices by using a multilevel model and including GP practices as a fixed effect, using a model which controls for the hierarchy of the data [37]. This could be attributed to patients from the same practice being more similar to each other, patient portals varying in functionalities, and knowledge and encouragement of their staff including GPs in using the patient portal [9].

One limitation in our study was that we could not measure or report the differences in patient portal functionalities that were included in the study, as our study included more than 1000 GP practices. Despite this, by including a large number of practices, we were able to analyze a more representative sample. We attempted to account for the differences between practices by running a multilevel model, recognizing that patients from the same GP practice may have more similar outcomes due to various factors including the type of patient portal offered in the practice and type of education and training available to both staff and patients [9].

While registration to the patient portal is the first step needed to benefit from its functionalities, studies show that without continuous or active use, patients cannot benefit from the technology. Registering to the portal does not guarantee continuous use of the portal, [38] which is essential to observe benefits associated with patient portal use [39]. A major limitation of this study was only using codes that indicated registration for accessing medical record as a proxy for patient portal use. This limited our study in two ways. First, access to medical records may have a minimal impact on prescription refill adherence. Second, patient portals offered through the NHS GPs, typically offer various functionalities beyond access to medical records, including online appointment booking and repeat prescription ordering [40]. However, as described in Multimedia Appendix 1, our search for codes that indicate patient portal use in CPRD Aurum only identified codes that indicate access to the medical records. This presented a challenge as patients using other portal functionalities, such as online appointment booking and prescription ordering, were not reflected in their electronic health records. Consequently, we could not determine if patients without these specific codes (codes in Multimedia Appendix 1) were indeed nonusers of patient portals and therefore, suitable as controls for our study. While we recognize that this limitation reduces the robustness of the results, it was the most suitable approach to examine this relationship that remains unexplored within the NHS, despite a wide availability of patient portals for nearly a decade. Further information about the use of patient portals such as prescription refill requests or appointment bookings made through the portal could better inform the relationship between patient portal use and medication adherence. Additionally, our study population included patients with chronic diseases including CVD, CKD, and diabetes, who may be less likely to use patient portals and more likely to enroll or register to patient portals [41]. According to a study conducted in the United States, patients with certain circulatory diseases such as heart failure, myocardial infraction, peripheral vascular disease, and those with diabetes and renal disease were less likely to use the patient portal of the health care system [42]. Given that patients included in this study included those with CVD, CKD, and diabetes, it is possible that although they were registered to the patient portal, these patients may not have been active users. Therefore, we cannot conclude whether the patient experienced reduced adherence to refilling their prescriptions due to the patient portal use based solely on the findings from this study.

In this study, patients had a slightly reduced adherence one year after patient portal registration, although adherence did not change after a 6 months follow up. This could be attributed to other mechanisms, including those mentioned above, that could influence medication adherence and were not included in this study. Registration to the portal alone does not influence health-related behaviors such as medication adherence. Future studies on recording of patient portal use in electronic health records such as log files of functionality use in patient portals (eg, appointment booking, messaging health care providers, or requesting medications) could enable studying health related outcomes and explain mechanisms involved between patient portal use and health outcomes.