This was a cohort study that included hypertensive patients, according to established criteria in national and international guidelines [1, 2, 11], who were 18 years or older and under stable antihypertensive treatment for at least 4 weeks. Participants performed a baseline HBPM, prescribed by their treating physician, between September, 1, 2008 and December, 31, 2015, in the Hypertension Section of Hospital Italiano de Buenos Aires. Duplicate HBPMs as well as HBPMs with less than 16 readings were excluded from the analysis.

The design of the study complied with the Code of Ethics of the World Medical Association (Declaration of Helsinki, 1964 and Declaration of Tokyo, 1975, as revised in 2008). The study protocol was approved by the local ethics committee. The patients duly authorized the use of the information in their medical records under the protection of their confidentiality through informed consent.

We used an automatic oscillometric device, Omron 705 CP (Omron® HEM-705CP-II, Omron®,Tokyo, Japan), previously validated [12] against a mercury sphygmomanometer according to the revised protocol of the British Hypertension Society [13], and appropriate cuff sizes according to each individual’s arm circumference. Patients received appropriate training to measure home BP after a 5-min rest, keeping their legs uncrossed, their back supported, and not talking. They registered duplicate sitting BP readings (1 min apart) in the non-dominant arm, during fixed hours in the morning (8–12 a.m.), afternoon (14–18 p.m.) and evening (20–24 p.m.), for four days. The reliability and reproducibility of this protocol of measurements has previously been addressed [14]. Briefly, in the cited study we assessed the reproducibility and reliability of a 4-day HBPM protocol with and without first-day measurements, analyzing a cohort of 353 subjects who required an HBPM for diagnostic purposes or evaluation of treatment efficacy. Reproducibility was quantified by test-re-test correlations and standard deviation of differences (SDD) between BP measurements obtained during the entire 4 days, with and without exclusion of the first day. The reliability criterion was the stabilization of the mean and standard deviation (SD). On the one hand, we found a strong test-re-test correlation between days 1 and 4 (0.80–0.91), which improved when we excluded the first day (p < 0.001). On the other, we found a reduction of the mean BP when we increased the number of days and a reduction of standard deviation of differences when we excluded day 1.

In the present study, morning readings were taken before breakfast and drug intake. The average of BP readings stored in the devices’ memory (not self–reported measurements) was used for analysis. According to current recommendations, first-day measurements were discarded [15, 16].

For the analysis, we considered a 4-day average of systolic and diastolic home BP, the average discarding first day measurements, and at each measurement period (morning, afternoon, and evening) separately. We also categorized home BP into adequate control when the average was <135/85 mmHg, and inadequate control if systolic BP was ≥135 and/or diastolic BP was ≥85 mmHg, for each of the five scenarios.

Our primary outcome was a composite of fatal and non-fatal cardiovascular events, including cardiovascular death, myocardial infarction, unstable angina, surgical and percutaneous coronary revascularization, congestive heart failure, atrial fibrillation, stroke (ischemic, hemorrhagic, or undetermined) and transient ischemic attack, occurring during follow-up. We also analyzed total mortality, cardiovascular mortality, and non-fatal cardiac and cerebrovascular events as secondary outcomes. Data regarding outcomes were obtained through the exhaustive manual review of each electronic health record, in all its modules: Ambulatory, Hospitalization, Emergency Room, and Home Hospitalization. The World Health Organization International Classification of Diseases (ICD-10), Volume 1, was used to codify the causes of death. Since patients who perform an HBPM in our hospital are affiliated to a prepaid medicine plan, they constitute a “captive” population, receiving healthcare only at the institution. This allows access to all follow-up data (without loss), except in the rare occasions when the prepaid plan cancellation occurs.

In all outcome analyses, we only considered the first event per participant within each category.

Medical records of all patients were reviewed to extract data regarding office BP level prior to HBPM, the type of antihypertensive drugs used at baseline, the presence of risk factors (diabetes, smoking status), and the history of cardiovascular disease (coronary heart disease and cerebrovascular disease). Laboratory data from 6 months prior to HBPM were also collected from medical records.

Regarding office BP, one to three measurements were taken after at least 5-min sitting rest using a standard validated aneroid sphygmomanometer (Riester®, Jungingen, Germany or Wellch-Allyn®, Amsterdam, The Netherlands) or a validated automated upper arm-cuff devices (Omron® HEM-705CP-II or Omron® 7 200, Omron®,Tokyo, Japan) and appropriate cuff sizes according to each individual’s arm circumference. The average BP of available readings was used in the analysis.

The sample size was estimated assuming an annual cardiovascular event rate of 2.5%. This figure was extracted from other cohorts that also evaluated hypertensive patients under treatment [9, 10]. For a mean follow-up of 6 years in our cohort, we expected to have 15 events per each 100 included subjects. According to Peduzzi et al., in order to ensure the accuracy and precision of estimated coefficients though Cox regression models, the number of events for each included independent variable must be at least 10 [17]. Given that we planned to include 13 co-variables, plus our main variable of interest -home BP- we needed to observe at least 140 events (14*10). Therefore, we had to include at least 934 patients (100*140/15).

Quantitative data are expressed as mean and standard deviation or median and interquartile range, according to data distribution. Qualitative data are expressed as absolute and relative frequency.

The prognostic value of home BP in terms of cardiovascular events was analyzed through Cox regression models (proportional hazards analysis), which accommodate censored data, estimating unadjusted and adjusted hazard ratios along with their 95% confidence interval. In the adjusted models, hazard ratios were adjusted for office systolic and diastolic BP, sex, age, body mass index, number of antihypertensive drugs, smoking habits, diabetes, history of cardiac and cerebrovascular disease, fasting plasma glucose, total cholesterol and creatinine level. Home BP was analyzed as a continuous and as a dichotomous variable (uncontrolled vs. controlled BP). We used the Akaike information criteria (AIC) to compare different modeling strategies. AIC are model selection criteria, i.e., statistical tools that help identify the best-fitted candidate model among a set of candidates. The best model is the one that obtains the lowest score, which measures how much the evaluated model deviates from a theoretical model that shows a perfect fit. To compare models, the AIC of each model is calculated. If a model is more than 2 AIC units lower than another, then it is considered significantly better than that model [18].

All hypothesis tests were two-tailed, and a p value <0.05 was considered statistically significant.