Enhanced hypertension care through private clinics in Pakistan: a cluster randomised trial

Study design, setting, and participants

The study protocol has been published previously.¹⁹ A two-arm cluster randomised controlled intervention trial was conducted at 26 selected private clinics (clusters), in three districts of Punjab province (Sargodha, Mandi-Bahauddin, and Kasur), during the period January 2015–September 2016. A cluster design was used primarily because the intervention was delivered by doctors who required training and resources to deliver the intervention, and it would not have been possible for doctors to effectively treat patients under usual care and/or control conditions once trained on the intervention processes and provided with the intervention resources.

In the three districts, 66 private clinics that had been already mapped, selected, and engaged (for ≥1 years) in the delivery of TB care were listed. All the listed clinics, when selected, were found qualified on the minimal eligibility criteria, that is ≥1 room with furniture and/or equipment, one doctor, and one clinic assistant, daily outpatient load ≥20, and willingness to participate in the partnership arrangement. Then 26 clinics were randomly selected, proportionately, from each district. The research staff informed each clinic in-charge (doctor) and asked for their consent to participate in the trial. After consent, the 26 clinics were randomised into intervention and control arms (1:1 basis). A designated doctor and allied staff at each of the 26 clinics were trained on standardised hypertension diagnosis and record-keeping; whereas at 13 intervention clinics, the designated doctor was trained on the standardised clinical care, and allied staff members were trained on lifestyle education and patient follow-up.

Patients aged ≥25 years were included in the trial after their eligibility was assessed and their (informed) consent to participate in the trial was obtained. Those excluded were pregnant patients, those with already known hypertension or secondary hypertension, those with advanced chronic disease, those on CVD treatment previously, or those who knew they would not be residing in the catchment area for the required follow-up period of 9 months. Every eligible patient attending the clinic, and consenting to participate, was enrolled until the required number was achieved in all clusters.

Randomisation and blinding

The 26 clinics in three districts (that is, n = 8, n =8, and n =10) were randomised to the intervention and control arm on a 1:1 basis using a lottery method. The number of clusters from each district were chosen proportionately, but randomisation was not stratified by district. From 26 sealed envelopes, each containing a clinic name, in the presence of trial steering committee members, a staff member of the provincial directorate randomly selected 13 envelopes for the intervention arm and 13 for the control arm. To avoid protocol contamination, the staff at the intervention and control clinics were trained separately. It was not possible to blind the healthcare staff and patients to the allocation of trial clinics owing to the nature of intervention. However, outcome assessors were blinded to clinic allocation and any previous clinical measurements for patients.

Package of care at intervention and control clinics

The doctors and allied staff in both cluster arms were given the same training on initial patient screening (for example, overweight), diagnosing hypertension, and using a chronic disease card for recording. For study purposes, both arms were provided with resources over and above what was available in usual care. These included provision of disease cards (to record data at baseline and follow-up), electronic sphygmomanometer, weighing scale, glucometer and glucose test strips, thiazide tablets, and vouchers for free testing of serum cholesterol at identified laboratories.

In the intervention arm, along with these resources, the enhanced management of hypertension and associated conditions was supported by providing: (a) a context-adapted desk guide for step-by-step clinical care, including adding drugs and amending dosage of drugs for hypertension and associated conditions; (b) contextualised pictorial tool to educate patients regarding hypertension care and lifestyle change; (c) training of doctors and 'paramedics' (qualified technical staff members with training in general medical care provision) to deliver care using the desk guide and education materials; and (d) mobile phone and pre-paid cards for the allied staff to message and/or call patients about follow-up visits.

Each registered patient was asked to visit the clinic on a monthly basis, or earlier if required for any medical reason. At each monthly follow-up visit, the patients were clinically assessed, prescribed, dispensed (only thiazide), educated, and their records were updated. At intervention clinics, all these care tasks were carried out as per recommended intervention protocols; whereas at control clinics, the care was delivered as per routine practice at the respective clinic. At intervention clinics, patient adherence to follow-up visits was also supported through mobile phone reminders as and when needed. To keep the trial evidence relevant to the real-life programme implementation processes, any special measure for an enhanced staff adherence to the intervention protocol was deliberately avoided. Staff and patient adherence, and challenges to the intervention protocols, have been studied and published in this journal,²⁰ and the guides and tools are available at http://comdis-hsd.leeds.ac.uk/resources/ncd-care-package/.

Data collection and management

The data were collected, as a part of care delivery process, on dedicated NCD record cards. In both arms, recruited patients’ baseline clinical data on examination (for example, BP, blood glucose, serum cholesterol, and HbA1c) and prescription (that is drugs and dosage) were recorded by the doctor, and non-clinical data (for example, age, contact details, height, and weight) were recorded by the allied staff. Patients were followed-up for an outcome measurement at 9 months of treatment. Those who were deceased or did not turn up for the 9-month follow-up for 3 months were declared as lost to follow-up.

Patient NCD card data were stored on an SPSS database (version 17) by trained data entry staff with regular quality checks by a supervisor.²¹

Outcome measures

All outcomes were measured at the individual-level. The primary treatment outcome was the mean change in SBP (mmHg) between baseline and 9 months. The secondary outcomes were: (a) mean change in DBP (mmHg), total serum cholesterol (mg/dL), and HbA1c (%); (b) 9-month control of SBP (≤140 mmHg), serum cholesterol (<200 mg/dL), and HbA1c (≤7%); (c) reported rate of smoking cessation among smokers; and (d) reported cardiovascular and cerebrovascular events.

Sample size

In the protocol sample size, it was initially planned for a fixed, recruitable total of 12 clusters, with 76 patients per cluster. However, in response to a slower than expected patient recruitment rate per cluster in the piloting phase, the number of clusters was increased to achieve the desired number of patients recruited. Therefore, the ultimate aim was to recruit 26 clusters (13 per arm), assuming an average of 35 patients would be recruited in each cluster with a 25% loss to follow-up (that is, 26 patients followed-up per cluster), giving a total of 912 patients recruited with 676 followed-up. Based on the original protocol assumptions of an outcome standard deviation of 11²² and an intra-cluster correlation coefficient of 0.02,²³ this gave 80% power (for a 5% level of significance and two-tailed test) to detect a difference in the mean change in SBP of ≥2.1 mmHg between the intervention and the control arm.^24,25

Statistical analysis

The methods used were those that were suitable for cluster randomised controlled trials having a relatively small number of clusters per arm.²⁶ To analyse the crude effect of the intervention on the primary outcome, cluster-level summary values were calculated, which were based on the mean outcome in each cluster. An independent t-test was then used to calculate the treatment effect as the mean difference in the primary outcome between treatment arms (control minus intervention), along with the associated P value and 95% CIs. A two-stage approach was then used to adjust for potentially confounding covariates. First, a linear regression model was fitted to the individual-level primary outcome data, which adjusted for the covariates of interest excluding the treatment effect. The covariates were chosen before analysis and included sex, years of education, age, SBP at baseline, serum cholesterol at baseline, HbA1c at baseline, and BMI at baseline. This model was then used to calculate a covariate-adjusted difference-residual for each cluster, by calculating the mean difference between the observed and model-predicted outcome for every cluster. These cluster-level difference residuals were then analysed using an independent t-test as described above to estimate the covariate-adjusted intervention effect.

The continuous secondary outcomes were analysed using the same methods as the primary outcome analysis. For the binary secondary outcomes, the cluster-level proportions were calculated first, and the crude difference in outcome proportions were estimated between treatment arms (control minus intervention; that is, the risk difference), along with the associated P values and 95% confidence intervals, using an independent t-test. To adjust for covariates, the same two-stage process as described was used, but instead using a logistic regression model for the individual-level outcome data to calculate cluster-level difference-residuals, which were then analysed using an independent t-test to estimate the covariate-adjusted intervention effect.²⁶

The crude and covariate-adjusted extent of effect modification due to sex on the primary outcome was estimated by calculating, for each cluster, the mean female-minus-male difference in the primary outcome (for the covariate-adjusted analysis, cluster-level difference residuals were used in place of raw means, calculated as explained above, but without controlling for sex). These crude and covariate-adjusted male-female difference values were then compared between treatment arms (control versus intervention) using an independent t-test to estimate the crude and covariate-adjusted between sex difference in the treatment effect.

Statistical significance was set at 5% and two-sided P values calculated. All data were analysed as per the original allocation of clusters, and when patients were missing outcome and/or covariate data, they were excluded from all relevant analyses (complete case analysis).