The feasibility of assessing prognosis over 3 years in persons with a previous stroke/transient ischemic attack in general practice

Abstract

Background

Stroke has devastating consequences for survivors. Hypertension is the most important modifiable risk factor, and its management largely takes place in primary care. However, most stroke-based research does not occur in this setting. Ongoing hypertension and a risk of further stroke are a major concern for both patients and their general practitioners. We aim to assess whether it is feasible to assess prognosis in persons, with a previous stroke or transient ischemic attack (TIA), in general practice, and whether a well-powered observational study is possible.

Methods

We performed a search of the electronic health record of individuals previously identified as having had a stroke or TIA, to assess prognosis over 3 years. Feasibility was assessed by meeting five criteria: (1) all general practices approached participated, (2) greater than 90% of patient records were accessible, (3) all study outcomes were available to review, (4) that collection data was less than 15 min per patient, and (5) a power calculation for a planned observational study could take place.

Results

All six general practices approached participated freely, and 193/196 patients’ files were reidentified (98.5%). Twenty-eight cardiovascular events were recorded—most commonly a repeat TIA or ischemic stroke. Data collection took on average 5.5 min per file, and a power calculation for a planned observational study was completed.

Conclusion

This study demonstrates that the proposed methodology for a full cohort study within general practice of patients post-stroke/TIA is both acceptable to practices and feasible. An adequately powered, “time-to-event” study is possible.

Key messages regarding feasibility

• What uncertainties existed regarding the feasibility?

It was uncertain whether the type of information required for a prognosis study was accessible via patients own general practice records

• What are the key feasibility findings?

All six general practices approached participated freely, and 193/196 patients’ files were reidentified (98.5%). Twenty-eight cardiovascular events were recorded—most commonly a repeat TIA or ischemic stroke, but all outcome data was easily accessible in patients own records. Data collection took on average 5.5 min per file, and a power calculation for a planned observational study was completed.

• What are the implications of the feasibility findings for the design of the main study?

This study demonstrates that the proposed methodology for a full cohort study within general practice of patients post-stroke/TIA is both acceptable to practices and feasible. An adequately powered, “time-to-event” study is possible

Introduction

Stroke has major implications for survivors. Cognitive, motor, and functional independence is often compromised, and this impedes an individuals’ ability to perform their activities of daily living [1]. In addition to the personal and social consequences of stroke, the economic burden individuals carry can also be frightening, mainly due to the increased healthcare costs involved in stroke care and the concurrent loss of salary. It is believed that the current global incidence of acute stroke is 12.2 million cases per year, and that 101 million survivors are currently living with post-stroke challenges [2]. Unfortunately, stroke is the second most common cause of death, and the leading cause of disability, in Europe [3]. In 2017, the total economic cost of stroke care in the European Union was reported to be 60 € billion. This figure includes the healthcare and social costs of stroke care but also includes the potential loss to the economy of persons who were previously working [4]. There are modifiable and non-modifiable risk factors associated with stroke. Non-modifiable risk factors include family history, age, sex, and ethnicity. Modifiable risk factors include elevated blood pressure (BP), body mass index (BMI), physical activity levels, lipid/glucose levels, and cigarette intake, while those with treated or undiagnosed atrial fibrillation are exposed to higher risk. While BP is a key modifiable risk factor—treating physicians should also mindful that a reduction in risk should involve a comprehensive modifiable risk factor plan [5]. Research has shown interventions that improve a persons’ modifiable risk factors have the potential to decrease the occurrence of stroke and transient ischemic attacks (TIAs) drastically [2, 5]. It is known that hypertension has complex interactions with the other modifiable risk factors. It is thus vital to control BP in this setting, as it is the most important thing a healthcare worker can do to prevent stroke or repeat stroke [2, 6, 7].

One of the major predictors for a second stroke is having a first stroke. Secondary strokes are predominately associated with increased disability; therefore, it is vital to introduce targeted management following an initial event to decrease the risk of further morbidity/mortality. The 5-year risk for a recurrent stroke is estimated at 9.5% [8]. Unfortunately, it has also been shown that behavior is hard to change, even when a particular behavior clearly contributes to poor health [9]. Patient and public involvement (PPI) is underutilized in stroke research, and if this approach is more widely adopted, it is possible that outcomes may improve [10].

Up to 70% of persons post-stroke have sub-optimal BP control [8, 9]. The reasons for sub-optimal BP control are multifactorial including therapeutic inertia amongst treating doctors to increase medication doses, sub-optimal adherence to therapy, and the difficulties with modifying challenging lifestyle behaviors [11]. Reducing BP by 5/2 mm Hg (systolic BP/diastolic BP) can reduce the first incidence of stroke by 29% [12], while the risk of recurrent stroke is reduced by 28%, if a dual antihypertensive agent strategy is pursued [13].

Complicating matters further for general practitioners is the fact that incongruity exists between European, United Kingdom, and American hypertension guideline producing agencies on what BP targets patients should achieve. The American Heart Association recommend BP lowering to < 130/80 mm Hg [14], while the UK, National Institute for Health and Care Excellence (guideline 136) recommends BP < 140/90 mm Hg [15]. The European Society of Cardiology also commits to the lowering of systolic BP to < 130 mm Hg in 2024 [16]—but in its 2019 guidelines, it said it was not certain of the evidence for this [17].

The primary aim of this study is to determine the feasibility of assessing prognosis in a cohort of patients with a diagnosis of stroke or TIA, previously identified in 2019 [18]. Feasibility will involve examining the ease of practice recruitment, the accessibility of patient records and outcomes, time spent accessing these records, and the ability to plan for a more detailed future study. While recent research has examined the factors that underpin the incidence of repeat stroke from secondary care perspectives [19], a core research gap exists when there is a lack of understanding, for any relevant condition, of “what exactly is going on in primary care?” [20], where the management of most patients’ BP occurs [21].

Methods

Design and setting

In 2019, a cross-sectional study examining the prevalence of BP control in a cohort of patients who have survived a previous stroke or TIA in Irish general practice was undertaken (involving 328 patients)—there were not any specific exclusion criteria [18]. All patients were from practices (n = 10) affiliated with University of Limerick Education and Research Network-General Practice (ULEARN-GP) [22]. The practices used either one of two practice software systems (Socrates®, HealthOne®) and the International Classification for Primary Care (ICPC-2) to code for disease on practice software systems. Within this study, we examine the records of 196 patients from 6 of these practices.

Irish healthcare system

Ireland does not have universal registration with a general practitioner. Most patients aged over 70 years, or less than 8 years, and those below defined income levels receive free primary care. These patients represent almost 50% of the population and are registered with the Primary Care Reimbursement Service (PCRS; http://www.hse.ie/eng/staff/PCRS/), with the remainder being described as private patients and can pay to see any general practitioner (GP).

Recruitment

For this feasibility study, 6 practices were pragmatically chosen from the original 10 to reflect practice diversity (small vs large vs mid-sized, urban vs rural vs mixed, single-handed practice vs multi-GP practice, patient electronic record software system). These practices were selected as they were a fair reflection of the various types of practices we might encounter in a more detailed prognosis study. All chosen practices were located within 1-h travel from the university and had a mix of PCRS and private patients.

Feasibility outcomes

To assess whether the study is feasible or not, we selected five a priori parameters. These were as follows: (1) all general practices approached about the study were recruited; (2) greater than 90% of patient records would be accessible; (3) all cardiovascular and non-cardiovascular outcomes, listed in the research protocol, were fully identifiable in patients’ electronic charts; (4) that the time required to examine the data was less than 15 min per patient; and (5) a power calculation for a planned observational study could take place.

We will apply stop/go criteria to the feasibility parameters (1–5), and if these criteria are not met—we would not proceed with a larger study. We would prefer that 100% of patients’ records were accessible—but the study team agreed not to proceed if less than 90% could be accessed—as it is generally accepted that > 20% loss to follow-up leads to serious threats to study validity [23]. We felt that if we were any slower than assessing four files an hour—we could not complete data extraction from a single practice in a single day.

Data collection

In each of the six practices, a manual review of each patient electronic healthcare record was performed by two researchers (E. R. & H. G.). Training was provided by P. H. (senior author) to E. R. and H. G. on how to use the practice software system (Socrates^® & HealthOne^®) and more specifically how to search for the data required. All data was collected when two researchers were present, and queries were discussed with a senior third person (P. H.). Training for data collection was provided by P. H. who has > 10 years’ clinical and research experience in using electronic healthcare files. Regular weekly feedback sessions were held with the field researchers.

Cardiovascular and non-cardiovascular outcomes

These included all cause deaths, cardiovascular (CVD) deaths, non-fatal myocardial infarction, non-fatal stroke or TIA, non-fatal new heart failure, or non-fatal newly diagnosed peripheral vascular disease. To optimally describe the multimorbid status of patients post-stroke or TIA who attend general practice, we also recorded data on hospitalizations (CVD and non-CVD) and total number of medications taken per person. As coding for diseases in Irish general practice has not been universally adopted, identification of cause of death, or morbidity, was primarily through hospital or physician notices in patient files.

Patients are described as having a new myocardial infraction if they had newly coded ICPC-2 codes K75 or K76 on their electronic file or were identified as having such inhospital discharge letters. The ICPC-2 coding system is primarily used in Irish general practice.

The same was true for new stroke or TIA if they have ICPC-2 codes K89, K90, and K91 or were identified as having a new stroke or TIA on hospital discharge letters, and also for cardiac failure K77 and for new onset peripheral vascular disease K92.

We also recorded the last eGFR on file. Patients are described as having chronic kidney disease (CKD) if an estimated glomerular filtration rate (eGFR) was less than 60 ml/min/1.73 m² and was recorded on two occasions.

Blood pressure was the last office-based reading recorded on file. Elevated BP is defined as clinic systolic BP greater than or equal to 140 mm Hg and/or diastolic BP greater than or equal to 90 mm Hg (National Institute for Health and Care Excellence (NICE) guideline 136) [15]. We adopted this pragmatic approach to BP control as outlined by NICE-UK, as reducing BP to < 140/90 mm Hg is the first essential goal, before healthcare practitioners begin to contemplate lower targets.

Hospital admissions and diagnoses were identified by discharge summaries. The number of medications that each patient was taking was assessed via repeat prescription data in the patient electronic file.

Sample size

In this feasibility study, six practices participated (n = 196 persons). We hoped that greater than 90% of 196 patient records were accessible, and this would allow us to proceed with a larger, “survival analysis” study. The 95% confidence interval for records being accessible is 84.7% and 93.7% with a margin of error of 4.5%. Therefore, assuming at least 90% of 196 patient records are accessible, we can estimate this to within 4.5% with 95% confidence.

Statistical plan for this study

The descriptive statistics for the study were calculations based on fractions/percentages of the 196 patients who were previously identified in these 6 practices as having a previous stroke or TIA and who had the various cardiovascular and non-cardiovascular outcomes, as previously outlined.

Planned observational study (if feasible)

Our main hypothesis, for a future observational study, is that a combination of CKD and/or poorly controlled BP (> 139/89 mm Hg) predicts sub-optimal “survival” in a cohort of patients with previously documented stroke or TIA. Data from this feasibility study will potentially allow authors to calculate the power required for a planned “survival analysis” or “time-to-event” study. By “time to event,” we mean not encountering any of the following “events” in the 3-year period of follow-up (all-cause death, non-fatal MI, non-fatal stroke, and non-fatal heart failure and peripheral vascular disease, CVD hospitalization).

Statistical plan for the planned observational study (if feasible)

Predictors of long-term “time-to-event” tables across 3 years of follow-up will be performed using the Wilcoxon test. These will show the cumulative proportion “not having an event” across 3 years, in conjunction with the following baseline demographic clinical characteristics/CVD risk factors: gender (male /female), polypharmacy (five or more medications/ < 5 medications), hypertension control (yes/no), diabetes (yes/no), CKD (yes/no), and CKD or poorly controlled BP or both (yes/no).

Kaplan-Meyer curves to demonstrate “survival” according to sex, and the various clinical characteristics (polypharmacy, BP control, diabetes, CKD, CKD with or without poorly controlled BP), will also be assembled. We aim to perform Cox proportional hazards survival analysis (hazard ratios), and these will be adjusted for age and sex. For all analyses, P-values less than 0.05 will be considered as significant. Complete case analysis will be used as the approach regarding missing data—but only where the data is deemed “missing not at random” [24]. We will use the Bonferroni correction to adjust for multiple comparisons [25]. All the statistical analyses will be performed with the statistical software SPSS version 28.0.

Results

In the previous cross-sectional study, 328 patients were identified as having had a previous stroke or TIA [16]. For details, see Table 1. For this feasibility study, 6 of the initial 10 practices were willing to participate (n = 196 persons) (Fig. 1). All practices who were recruited agreed to participate.

Table 1 Characteristics of patients with a previous stroke/TIA in original cross-sectional group (n = 328)

Fig. 1

Flow chart of patients. TIA, transient ischemic attack; CVD, cardiovascular disease

In total, 193 patient records were successfully reidentified at the 6 practices (98.5%)—and data was collected between December 2021 and July 2022. Three patients were likely mis-coded for stroke/TIA at the time of the initial study, and this likely precluded their reidentification [18]. A further 11 patients had transferred between general practices in the immediate period following the initial study in 2020, and data was not available on these persons either (Fig. 1).

During the search of electronic records—it was ascertained that 25 patients had died (13%), and 21 persons had non-fatal cardiovascular outcomes (11%), but a total of 28 individual non-fatal CVD events were recorded (Fig. 1). A composite outcome was also assessed which included (1) all deaths, (2) all cardiovascular outcome events, and (3) all cardiovascular admissions (not related to a specific cardiovascular outcomes). In total, 57 persons had the composite outcome. All outcomes were easily identifiable within patients’ electronic records.

A total of 18 h was spent at the practices for data collection, which was an average of 5.5 min per patient. There were no associated costs, aside from travel to and from practices to collect the data.

From data examined in this study, we showed that 106 participants with “CKD and/or poorly controlled BP” had 29 events, and 90 participants “without CKD or poorly controlled BP” had 17 events, using a significance level of 0.05, and a power of 0.8, and where the hazard ratio (or relative risk) is 1.5, an adequately powered study needs a total of 192 events and 428 study participants to succeed (https://sample-size.net/sample-size-survival-analysis/). The initial cross-sectional study in 2019 (of patients post-stroke/TIA) included 328 persons from 10 practices—thus, the recruitment of an extra two large practices to the study will ensure adequate power in the proposed “time-to-event” study [18].

A more detailed examination of the studies data showed that the average age of patients from the six practices was 74.6 years. Overall, 60% of the population (n = 115) studied were male, and 89% of patients were registered with PCRS. Comorbidity was relatively common with 30% (n = 59) of patients having CKD and 14% having concurrent diabetes mellitus (n = 28). The mean number of medications on repeat prescription in this cohort was 8.6 medication per person. When BP readings were compared to the NICE clinical guideline-136 [15], the proportion of persons with controlled BP (< 140/90 mm Hg) improved from 65% (n = 127) to 76% (n = 149) over the 3-year period. Full details can be found in Table 2.

Table 2 Characteristics of patients with a previous stroke/TIA in original cross-sectional group (n = 193)

Unfortunately, 13% of study participants died over 3 years (n = 25), and 20% of these deaths were due to cardiovascular events (n = 5). A total of 21 patients suffered 28 non-fatal CVD events—for full details, see Table 3. In all, 14.4% of these events (n = 10) occurred in persons with their original BP recorded as ≥ 140/90 mm Hg (n = 69), and 12.5% (n = 16) occurred in those who had their BP controlled (n = 127). Similarly, a repeat stroke/TIA was more commonly observed in those who had uncontrolled BP compared with those who are normotensive 5.8% (4/69) vs 4.7% (6/127).

Table 3 Outcome events for patients’ post-stroke or TIA in a feasibility study (n = 193)

It was determined that 38% (n = 75) of patients were hospitalized for any reason between 2019 and 2022. Thirty-one of these admissions were for cardiovascular-related hospitalizations and the remainder for various other indications (Table 4).

Table 4 Hospitalization events for patients’ post-stroke or TIA in a feasibility study (n = 193)

Discussion

Prior to attempting a larger “time-to-event” study, it was necessary to demonstrate that our proposed methodology was feasible by discovering and addressing key uncertainties. These uncertainties included the acceptability of a follow-up study to practices (recruitment), the accessibility of patients’ files and outcome data, the workload involved for the researchers, and whether this study could inform a sample size calculation for a fully powered observational study. As all five of these key uncertainties were addressed, the planned observational (time to event) study is possible.

To ensure a smooth data collection process, E. R. and H. G. contacted the practices in advance of arrival to schedule a convenient time to access the practice computer. Adequate training of the data collectors was provided by P. H. prior to entering the practices this helped to improve time management and efficiency when there. This study appeared to place little burden on any of the participating practices teams. Issues to address in a larger “time-to-event” study include signing individual practice confidentiality agreements to further allay any fears general practitioners may have about data processing and organizing a brief virtual meeting with practice administrators prior to attending the clinic to orientate researchers to the specific practice software nuances.

Interestingly, the study suggested that 22 further persons had their BP normalized during the 3 years of follow-up. During this period (2019–2022), the Health Service Executive (HSE) in Ireland commenced the chronic disease management program in conjunction with the state’s general practitioners (https://www.hse.ie/eng/about/who/gmscontracts/2019agreement/chronic-disease-management-programme/). This involves specific payments to GPs for identifying patients post-stroke/TIA in their practices and engaging these patients in a structured treatment and prevention program. This potential association requires further examination in studies with control subjects.

There is a paucity of research examining the management of patients care post-stroke in general practice despite a 2015 national stroke audit recommending health authorities address this issue by improving funding and resources for this type project [26]. A national study conducted in Ireland has previously concluded that general practices were not accomplishing goals in relation to stroke prevention [27]. Interestingly, BP control rates in this population post-stroke/TIA in Ireland are higher than that of Norway, where 47% of those enrolled in a prospective cohort study of those post-stroke had their BP adequately controlled (< 140/90 mm Hg) [28]. While this study is situated in a European union member state, there is an urgent need to improve stroke care in lower- and middle-income countries (LMIC) [29]. There is evidence that persons who seek healthcare post-stroke in LMICs tend to be those with more severe strokes rather than those with minor strokes who would benefit from urgent review and repeat stroke risk factor reduction. These issues are more prevalent in healthcare systems without universal care or where cost is a key prohibitive factor in attending a doctor [30].

The very fact is that over a third of patients in this cohort (38%) required hospitalization at some point during the study, and that the typical study participant takes eight or more medications daily suggests that this cohort is a high-risk, older, multi-morbid grouping. Interestingly, this figure tallies with a 10-year retrospective study on persons post-stroke, from Norway (n = 243), when at 30 months circa 45% of patients had required hospitalization (for any reason) [31]. Further studies may well clarify if some of these hospitalizations could be preventable with extra community resourcing and integrated primary-secondary care, community programs. Despite advances in stroke care, long-term prognosis remains a cause for concern. Data suggests that two-thirds of patients who have had an ischemic stroke, and three-quarters of patients who have had a hemorrhagic stroke were dead or dependent after 5 years had passed. There is a sparsity of large studies examining prognosis post-stroke [32].

Strengths and limitations

We believe this is the first assessment of prognosis in a cohort of patients with a previous stroke or TIA identified in Irish primary care, and we have shown that it is feasible to do this type of study.

The first limitation to note is that we selected practices in a nonrandom fashion for the study. Instead, it was convenience sample we used, based upon practice location, and the ability for practices to allow access to researchers’ access to data during the COVID-19 pandemic. However, a variety of practice types and locations (rural vs urban) were utilized (Appendix Table 5), and we do not believe that these chosen practices are particularly different to any of the nonselected practices.

Secondly, we did not follow-up on the patients whose files we could not access or those who had transferred to a different practice. Evidently, this will be a challenge in a larger-scale study, as these figures would be expected to increase with a larger number of practices involved. Resources and extra consideration as to how to tackle this issue should be put in place for any follow-up studies. This may involve directly contacting patients or their families for details on prognosis and would require appropriate ethical approval.

We acknowledge that the manual extraction of data can lead to potential data errors between the patient file and the study spreadsheet. We mitigated this by working in pairs during data extraction, double-checking unusual/incomplete data before the data extraction moved to the next study participant, and adequate training for researchers in the use of electronic healthcare files. A third senior author (P. H.) was also available to answer data extraction queries at any stage while the process was ongoing.

We also accept that post-stroke care should attempt to fully mitigate against all potential risk factors to prevent future stroke, and this should include measures to improve lipids, physical activity levels, smoking status, and blood sugars/body mass index, as well as improving BP.

Finally, in Ireland, there is a mixture of both private and public providers in the health system. It is possible that important outcome data may be missing from patient files, as private patients can technically attend any GP they wish to see, and researchers may not be aware of these consultations.

Conclusions

This study demonstrates that the proposed methodology for a well-powered observational study within general practice, involving patients post-stroke/TIA, is both acceptable to practices and feasible. An adequately powered, “time-to-event” study appears possible.

Appendix

Table 5 Practice table