Seniors-ENRICA-2is a cohort of 3273 community-dwelling individuals, aged 65–94 years, residing in Madrid, selected in January 2016–December 2017 by sex- and district-stratified random sampling of all individuals holding a national healthcare card. Since all people residing in Spain are entitled to free healthcare, the list of cardholders closely approximates the entire resident population of Madrid.20 Data were collected in 3 stages by personnel trained in the study procedures: a phone interview to obtain sociodemographic, lifestyle, and morbidity data; a home visit to collect blood samples; and another home visit to perform physical examination (including BP measurements).

As derivation sample to calculate the IAH score, we used data from 2394 subjects who accepted to undergo ABPM and had successful readings. As external-validation sample we used data from the Seniors-ENRICA-1 cohort, comprising 1047 community-dwelling participants aged≥60 years across Spain, assessed in October 2012 to October 2013 following similar data collection to Seniors-ENRICA-2 cohort, and with valid ABPM as done by other studies.2,21,22

Study participants provided written informed consent, and the Clinical Research Ethics Committee of Hospital Universitario La Paz, Madrid, approved the Seniors-ENRICA-2 and Seniors- ENRICA-1 cohort-studies. This manuscript follows the recommendations of the TRIPOD statement (Table 1 of the supplementary data).23

We used available data on a number of variables associated with MH,1–4,6,10,11,17 and also used opinion of experts in MH in our research team. First, we considered sex, age, and self-reported health behaviors, including tobacco smoking (current-, ex- or never-smokers), sleep quality (considering “poor sleep” if reporting “almost always or sometimes difficulty falling asleep or awakening during the night” or “very bad, bad or regular sleep quality”), sleep duration (“how many hours do you usually sleep per day including both nighttime and daytime?”), physical activity (time in hours per week that they spend walking, including commuting), and sedentary behavior (being seated more than 8h a day, including sitting time watching TV, on the computer, listening to music, reading, during meals or during transportation). Diabetes was approached as a physician diagnosis, current use of glucose-lowering drugs, or fasting serum glucose >125mg/dL; chronic kidney disease as glomerular filtration rate <60mL/min/1.73m2 as per Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation. Sleep apnea syndrome and depression requiring treatment were also considered, based on self-report of a physician diagnosis. Body mass index (BMI) was calculated as measured weight in kilograms divided by squared measured height in meters (overweight, BMI 25–29.9kg/m2; obesity, ≥30kg/m2).

Single-point casual BP was measured 3 times (at 2-min intervals, after resting 5min in a seated position) at 1 visit, by certified examiners, following standardized conditions,8,22 using a validated oscillometric device (Mobil-O-Graph 24h PWA monitor, I.E.M., Germany; Mediscan, Spain) and appropriate-size cuffs.

Thereafter, 24-h ABPM was performed using the Mobil-O-Graph device, which registers BP at 20-min intervals for the 24-h period. The ABPM recordings were performed on working days and the participants were instructed to maintain their usual activities, and to keep the arm extended and immobile at the time of each cuff inflation (appropriate-size cuffs were used). ABPM recordings were considered valid if ≥70% BP measurements were successful during both the daytime and nocturnal periods, defined individually according to the patient's self-reported time of going to bed and getting up.7

IAH was defined as casual home BP<140/90mmHg (mean of the second and third measurements) and mean 24-h ABPM≥130/80mmHg in the absence of antihypertensive drug treatment.7,8

For the derivation sample, 1331 participants were excluded for being on antihypertensive treatment, 375 for having elevated casual BP and 43 for lacking data for variables; thus, the analyses were performed with 645 untreated subjects with casual BP (mean of the last 2 readings) <140/90mmHg. For the external validation sample, after applying the same exclusion criteria, the analyses were performed with 356 untreated subjects with casual BP <140/90mmHg (mean of the last 2 BP). Between-group differences in clinical characteristics were assessed with the Chi-square test for categorical variables and the Student t test for continuous variables.

For the derivation sample, the association between each clinical characteristic and the presence of IAH was summarized with odds ratios (OR), and 95% confidence intervals (95%CI), obtained from logistic regression. Variables that attained univariate statistical significance (P<.1), were included in a multivariable model, and a forward stepwise selection procedure was used. The final model also included sex and age. As sensitivity analysis, we built a model adding interactions of sex and age with the other covariates and tested if model prediction based on the Akaike information criterion and the area under the receiver operating characteristic curve (AUC) was improved. Beta coefficients from the multivariable model were used to assign points to each variable in a score (range, 0–12) created to estimate the probability of having IAH. Variables that were associated with IAH were usually assigned 1 point for simplicity of use of the score, while those with stronger association were given 2–4 points, according to widely used procedures.24,25 We stratified the risk of having IAH into 3 categories: high (risk>50%), middle (risk>25%), and low (<25%).

To assess the extent to which the IAH score discriminated between individuals with and without IAH, we calculated and plotted the AUC (and 95%CI). To assess the model calibration, we plotted the estimated and observed probability of IAH for each score point and used the Hosmer–Lemeshow's goodness-of-fit test (built on each risk-point) to statistically assess the differences between such probabilities. Since the cut-off points defining categories in the IAH score were somewhat arbitrary and can result in loss of information, we conducted a sensitivity analysis by replicating the IAH model using continuous categories of the same covariates, and again we estimated the AUC and used the Hosmer–Lemeshow test. Lastly, the diagnostic ability and calibration of the score developed from the derivation sample were assessed in the external-validation sample.

The size-efficiency of performing ABPM according to the subjects’ IAH score, i.e., the number of subjects at a given risk (or threshold) score needed to be examined with ABPM to diagnose 1 case of IAH (NND), was calculated as follows: 1/probability of IAH at each risk point (IAHi) or at and above each risk point (IAHi+). Probability of IAHi+ was calculated as Σi([Ni/Ni+]*probability of IAHi), where Ni and Ni+ stand for the number of individuals at each point and at and above each point, respectively. By convention, we rounded any decimal NND upwards to the next whole number.25 Confidence intervals were corrected for small samples where appropriate.

Analyses were conducted in April–September 2020, using Stata/SE version 13 (Stata Corp, United States), SPSS version 24 (IBM, United States), and R version 3.0.2 (R Foundation for Statistical Computing, Austria).

In the multivariable analysis, the variables significantly associated with IAH were: BMI 25–29kg/m2 (OR, 2.58; 95%CI, 1.49–4.45), BMI≥30kg/m2 (OR, 3.53; 95%CI, 1.79–6.95), the first casual BP≥140/90mmHg (OR 2.49; 95%CI, 1.53–4.04), and the mean of the second and third casual BPs within the “120–9/80–4mmHg” category (OR, 2.52; 95%CI, 1.16–5.51) or within the “130–9/85–9 mmHg” category (OR, 7.70; 95%CI, 3.71–15.99) (Table 2). OR for age≥80 years was 2.98 (95%CI, 0.97–9.17; P=.05), and 1.30 for male sex (95%CI, 0.83–2.03; P=.2).

Score points assigned to the variables associated with IAH were as follows: men, 1 point; age≥80 years, 2 points; BMI of 25–29, 2 points; BMI≥30kg/m2, 3 points; first BP measurement≥140/90, 2 points; mean of the second and third BP measurements of 120–9/80–4, 2 points; and 4 points if this mean was 130–9/85–9mmHg (Fig. 1). The probability of IAH was 0.02, 0.03, 0.05, 0.08, 0.12, 0.18, 0.25, 0.35, 0.47, 0.59, and 0.72 for scores of 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, and ≥10, respectively (Fig. 1).

Fig. 1.

Probability of having isolated ambulatory hypertension based on the score estimated from the multivariable logistic model, among normotensive subjects aged 65 and over.

(0.2MB).

The AUC for the score was 0.80; 95%CI, 0.77–0.84 (Fig. 1A of the supplementary data), and the Hosmer–Lemeshow test P value was 0.18 for the observed vs predicted IAH probabilities, indicating support for a properly calibrated model (Fig. 2A).

Fig. 2.

Calibration of the isolated ambulatory hypertension (IAH) score. Observed (A) and expects (B): observed and model predicted prevalence of isolated ambulatory hypertension, respectively. X axis: risk-score points of IAH; Y axis: percentage of individuals with IAH.

(0.18MB).

Figs. 2 and 3 of the supplementary data show the AUC of models without and with only the variable most strongly associated with IAH (“mean of the last 2 BP”; AUC, 0.75 and 0.73, respectively). Model prediction based on the AIC slightly improved after consideration of interaction of sex and age with other covariates, from 127 to 115 (Table 2 of the supplementary data). Likewise, model's AUC was only slightly higher when interaction terms were added (0.81–0.83). Lastly, when all variables in the multivariable model were modeled as continuous, the AUC was 0.80 (0.76–0.84) (Fig. 4 of the supplementary data), practically the same as the AUC of the main model.

In the external validation sample, probabilities of IAH were 0.02–0.36 for scores of 0–≥10 (Fig. 5 of the supplementary data). The AUC for the score was 0.73; 95%CI, 0.64–0.82 (Fig. 1B of the supplementary data), and the Hosmer–Lemeshow's P-value was .93 for the observed vs predicted IAH probabilities (Fig. 2B).

The NNDs for a score of 6, 7, 8, 9, and ≥10 points were 4, 3, 3, 2, and 2, respectively (Fig. 3, column 4, orange for the 3 first NNDs and green for the last 2).

Fig. 3.

Probability of having IAH and number of subjects needed to be examined with ambulatory blood-pressure monitoring to detect one with IAH, according to the risk score of IAH. Untreated normotensive subjects aged 65 and over. NNDi: NND at each risk point (i). NNDi+: NND at (i) and above a given risk point (i+). Specific risk-point metrics for physicians: high-risk status of having IAH (red) defined as probability >50% (≥9 points); middle-risk status (orange) defined as risk >25% (≥6 points). Cumulative risk-threshold metrics for healthcare managers: high-risk status (red) set at a risk threshold of 50% (≥8 risk points); middle-risk status (orange) set at a threshold of 25% (≥3 risk points). By convention, any decimal NND is rounded upwards to the next whole number. IAH, isolated ambulatory hypertension; NND; number needed to detect.

(0.36MB).

In cumulative terms, 12.4% of all the subjects were at high risk of having IAH (≥8 points; Fig. 3, column 6), and 60.3% were at middle risk (>2 and <8 points). The NNDs for a score of ≥8 points (high probability of IAH, Fig. 3, column 7 in red) and ≥3 points (middle risk, orange) were 2 (Fig. 3, column 8 in green) and 3–4 (orange), respectively.

In the external validation sample, NNDs for scores of 0–≥10 were 50, 50, 34, 25, 17, 13, 10, 7, 5, 4, and 3, respectively (Fig. 5 of the supplementary data). NNDs for middle-risk score (≥6 points) was 6, and it was 4 for high-risk subjects (>8 points).

The screening score may help the physician and the healthcare manager to decide on the appropriate use of ABPM. For the physician, individual risk of having IAH (or MH) would be easy to calculate, and ABPM could be a size-efficient strategy to detect IAH based on the score. In an individual with low probability of IAH (<25% or <6 points), avoiding ABPM could be a reasonable choice as it is low-risk for the patient and size-inefficient (Fig. 3, column 4, red). Conversely, reserving ABPM to patients with a high probability of IAH (>50% or >8 points), would be very size-efficient (only 2 patients would need ABPM to diagnose 1 with IAH).

For the healthcare manager, a policy oriented to considering ABPM in subjects with a probability of IAH>50% (at least 8 points) entails a population burden probably assumable by the health system, involving only 12% of all subjects in the population (Fig. 3, column 6) and being very size-efficient (only 2 subjects would need to be examined with ABPM to detect 1 with IAH; Fig. 3, column 8, green). The issue of a wider use of ABPM, including patients at middle risk of IAH (>25%), is one of affordability and would involve a burden of 60% of all the patients in our study population; nevertheless, only 3–4 of these subjects are needed to undergo ABPM to detect one with IAH.

This study presents some limitations. First, casual BP was not measured in the clinic but in subjects’ homes. Nevertheless, unlike self-measured home BP along several days, only 3 BP readings were taken on a single occasion by trained observers unfamiliar to the subjects, thus an observer effect and white-coat effect was present.29 These observer-measured home BP measurements have also been used in other studies to estimate MH and its consequences,21,29,30 and are increasingly used in clinical practice with many patients providing their doctors with BP taken at home either by the patient or another person. However, observer-measured home-BP may still be somewhat lower than in office where a clinical effect (setting) is added to the observer effect (staff),29 which could lead to some overestimation of the prevalence of MH on clinic BP. Nevertheless, BP-thresholds based on office-BP and home-BP measurement are similar in the lower BP range (including both 120/80 and 130/80mmHg),28 so the magnitude of the bias should be small. Also, IAH only include real (ambulatory) hypertension masked at home, thus likely losing some non-hypertensive patients in the clinic with ambulatory hypertension.

Second, ABPM data were obtained from a single ABPM recording, and reproducibility of MH is fair only in the short-term.1,13,31,32 Nevertheless, we only aimed to estimate the probability of having IAH at a moment in time – a single entire day-, and only 1 24-h ABPM is common clinical practice.1,17