Spousal concordance for hypertension: A meta‐analysis of observational studies

Abstract

The authors performed a meta‐analysis of observational studies to estimate the magnitude of spousal concordance for hypertension and to examine whether the concordance varied by important study methodological aspects. PubMed and Embase were searched up to June 2017 for cross‐sectional, case‐control, and cohort studies that investigated the concordance/association of hypertension between spouse pairs. A meta‐analysis with random‐effects models was performed by pooling adjusted odds ratios. Eight studies with a total number of 81 928 spouse pairs were eligible. The pooled results showed that spouses of individuals with hypertension had 41% (odds ratio, 1.41; 95% confidence interval, 1.21–1.64) increased odds of having hypertension themselves. The association applied to both men and women, and was not significantly different between studies with adjustment for body mass index and those without. The findings highlighted the importance of environmental factors in the development of hypertension.

1. INTRODUCTION

Hypertension is the leading risk factor for morbidities and mortalities caused by cardiovascular and kidney diseases. The Global Burden Project estimated that hypertension accounts for 9.4 million deaths each year worldwide.1 Hypertension is a complex disorder resulting from both genetic and environmental determinants.2 To date, only several rare Mendelian hypertensive phenotypes have been studied and the cumulative effects of all identified genetic loci on blood pressure (BP) variation remain small.3, 4 On the other hand, environmental and lifestyle exposures such as obesity, smoking, and excess salt intake have been suggested to be highly influential.5 Traditionally, the twin pair study design was frequently used to investigate the contribution of the genetic and environmental factors to diseases.6 Another important resource for such research purpose was spouse pairs. Spouses are generally genetically unrelated but share similar or common social context, economic positions, and living habits, providing an optimum resource for investigating environmental basis for diseases such as hypertension. If spouses of patients with hypertension are at a higher risk of hypertension themselves, then hypertension screening and prevention strategies could be tailored to couples rather than individuals in order to make greater achievements, which has important implications for public health practitioners and healthcare providers.

Although a few published studies have suggested a positive association of hypertension between spouses, the magnitudes of association differ widely, possibly because of the differences in statistical adjustment or ascertainment methods of outcomes.7, 8, 9, 10, 11, 12, 13, 14 For example, one earlier case‐control study performed in the United Kingdom demonstrated that among various risk factors including age, smoking, and body mass index (BMI), spousal diagnosis of hypertension was the strongest risk factor of hypertension for men and second strongest risk factor for women, with adjusted risk estimates at two‐ to three‐fold.7 In another community‐based population study in Brazil, however, the spousal association was not significant after adjusting for age.9 Therefore, these inconsistencies and uncertainties need to be addressed using quantitative and comparative methods.

Previous meta‐analyses have observed a significant positive spousal concordance for diabetes mellitus and the majority of main coronary risk factors.15, 16 However, no meta‐analysis has summarized the available evidence of spousal concordance for hypertension. In addition, the mechanism through which spousal concordance operates was not known. The concordance may operate through the effects of assortative mating or through cohabitation, which could be determined by examining whether the associations were influenced by adjustment for BMI (a surrogate for assortative mating) and the length of marriage. Moreover, different methods of outcome ascertainment may also affect the association. To address these questions, we performed a meta‐analysis of observational studies to estimate the magnitude of spousal concordance for hypertension and to examine whether the concordance varied by important methodological aspects such as adjustment for BMI and ascertainment of outcome.

2. METHODS

We conducted this meta‐analysis in accordance with Meta‐analysis of Observational Studies in Epidemiology (MOOSE) guidelines.17

2.1. Search strategy

We searched PubMed and Embase up to June 1, 2017, for original studies that investigated the concordance or associations of hypertension between pairs of spouses. The following search terms were used: (1) “spouses” OR “spousal” OR “couples” OR “partners” OR “married” OR “husbands” OR “wives” AND (2) “hypertension” OR “hypertensive” OR “blood pressure” AND (3) “odds ratio” OR “risk” OR “hazard ratio” OR “concordance” OR “aggregation” OR “correlation.” Corresponding medical subject headings were searched together with free text searching in titles and abstracts. Certain terms were modified to suit different databases. Details of the search syntax in PubMed and Embase are shown in Tables S1 and S2. Language restriction was not imposed. Therefore, we were able to include studies of all languages in the meta‐analysis with the assistance of language translators. We also manually examined the reference lists of important reviews and included studies for potential eligible studies.

2.2. Inclusion criteria and study selection

Studies meeting the following criteria were included: (1) cross‐sectional, case‐control or cohort design; (2) study sample of couples identified from survey or public health records, or administrative or hospital databases; (3) the outcome was hypertension, either ascertained by self‐reported history, physician's diagnosis, or direct measurements of BP, and the exposure was spousal hypertension status; and (4) quantitative concordance measurements were reported as adjusted or unadjusted odds ratios (ORs). We excluded studies that did not address with‐in spousal association of hypertension and those that only reported linear correlations of BPs. Hypertension was defined by self‐report or physician's diagnosis in some studies (previously diagnosed), and in other studies was defined by previous diagnosis in combination with direct BP measurements (previously undiagnosed). To extend the external validity of the results, we included studies that applied both definitions only if the outcome ascertainment was adequate and clinically valid.

Following database searches, two authors (ZW and YD) independently conducted the study selection process. Any disagreement was resolved by discussion or by consulting a third author (WJ). The first process of selection was based on screening of titles and abstracts to discard studies that were clearly irrelevant. Then, the full texts of remaining articles were reviewed to assess for eligibility in accordance with the inclusion criteria. If multiple publications were linked with the same study population, the one with a more comprehensive analysis was retained.

2.3. Data extraction

The outcome was hypertension and the exposure was statues of spousal hypertension. Two authors (ZW and YD) independently abstracted data from the included studies using a predesigned form. Relevant data included name of first author, study design, year of publication, country of origin, source of populations, age range and sex proportions, number of spouse pairs, ascertainment of outcomes, and confounders that were adjusted for. The methods of hypertension definition could be either: (1) self‐reported or physician's diagnosis of hypertension (diagnosed), or (2) self‐reported or physician's diagnosis of hypertension in combination with direct BP measurements (diagnosed and undiagnosed).

2.4. Data synthesis and statistical analyses

Studies generally presented ORs with different levels of adjustment. In our main analysis, we pooled the results with the most extensively adjusted ORs of each study. Most studies used women's hypertension as the outcome variable and husbands’ hypertension status as the exposure variable; therefore, we used ORs for women's hypertension in the main analysis and a secondary analysis was conducted using men's hypertension as outcome. We applied DerSimonian and Laird random‐effects models for meta‐analysis because this method accounts for both within‐study and between‐study variability and gives conservative effect estimates. Heterogeneity between studies was evaluated using the Cochrane Q and I ² statistics. Publication bias was assessed by visual inspection of funnel plot symmetry and by the Egger test and Begg test.18, 19

We also conducted a series of sensitivity and subgroup analyses to examine the robustness of results and the potential modifying factors of the association. First, we performed sensitivity analyses by removing one study per time and recalculating the pooled ORs of the remaining studies to identify studies that had a major influence on the results. Second, to capture associations likely mediated through assortative mating, we compared the ORs that adjusted for BMI (assuming BMI was a surrogate for assortative mating) with those that did not. We did this by separately pooling ORs in models that adjusted for BMI and ORs in models that adjusted for other variables (such as age and/or socioeconomic factors) but not BMI. Moreover, we also investigated whether the associations varied by different methods of hypertension ascertainment, by comparing ORs that applied self‐reported/physicians’ diagnosis of hypertension (diagnosed) with that using self‐reported/physicians’ diagnosis combined with direct BP measurements (diagnosed and undiagnosed).

All statistical analyses were performed using STATA 12.0 (StataCorp).

3. RESULTS

3.1. Identification of study and study characteristics

We retrieved 1259 and 1672 records from PubMed and Embase, respectively, from which 2189 unique records were identified and used for title and abstract screening. After screening, 24 articles were retained for a further step of thorough full‐text evaluation. Finally, eight studies fulfilled our inclusion criteria. The process of study selection was shown in Figure 1.

Figure 1.

Flow diagram of study selection

The Table summarizes general characteristics of the eight included studies. The numbers of spouse pairs varied considerably, ranging from 174 to 66 130. One of the studies was published in Russian, which we translated into English with the assistance of a language translator.12 The eight studies also differed in study design: one study was a case‐control,7 five were cross‐sectional,8, 9, 10, 11, 12 one was a cross‐sectional analysis of baseline data from a prospective cohort,14 and the last was a cross‐sectional analysis of a cumulative 9‐year longitudinal cohort.13 Regarding the countries of origin, two studies were conducted in the United Kingdom,7, 8 two in the United States,10, 13 two in Russia,12, 14 one in Brazil9 and one in China.11 The outcome was defined as diagnosed hypertension in four studies7, 8, 10, 11 while as both diagnosed and undiagnosed hypertension in another four studies.9, 12, 13, 14 Moreover, BMI was adjusted in four studies.7, 8, 10, 13

Table 1.

Characteristics of the eight included studies

Author, year	Study design	Country	Age range; mean age (y, SD)	Source of population	No. of spouse pairs	Ascertainment of hypertension	Mostly adjusted covariates
Hippisley‐Cox, 1998	Case‐control	United Kingdom	≥30; Women 56.7 (16.1), men 55.0 (15.0)	Computerized registered patients from general practice	1393	Diagnosed hypertension from registered computer database	Age, BMI, diabetes mellitus, and BP
Hippisley‐Cox, 2002	Cross‐sectional	United Kingdom	30–74	10 Practices from the Trent Focus Collaborative Research Network	8386	Read code for hypertension in computerized database	Age, smoking, and BMI
Bloch, 2003	Cross‐sectional	Brazil	≥20; Women 44.7, men 48.7	Community‐based population	365	Taking antihypertensive treatment or SBP ≥160 mm Hg or DBP ≥95 mm Hg	Age
Stimpson, 2005	Cross‐sectional	United States	65–94; Women 70.9 (5.2), men 73.9 (6.3)	Representative population from southwestern United States	553	Self‐reported diagnosis by physicians	Age, education, US nativity, BP, BMI, smoking, and drinking
Jurj, 2006	Cross‐sectional	China	40–70; Women 51.9 (8.8), men 54.6 (9.7)	Community‐based population	66 130	Self‐reported hypertension for women and wife‐reported hypertension for husbands	Age, occupation, education, and family income
Konnov, 2010	Cross‐sectional	Russia	32–63	Patients with premature CHD and their spouses	174	Discharge records or taking antihypertensive treatment or SBP ≥140 mm Hg or DBP ≥90 mm Hg	Age and sex
McAdams DeMarco, 2011	Cross‐sectional analysis of longitudinal cohort	United States	45–64; Women 53, men 55	Community‐based population	4500	Self‐reported antihypertensive treatment or SBP ≥140 or DBP ≥90 mm Hg	Age, race, BMI, smoking, and sodium intake of both partners
Dolgalev, 2013	Cross‐sectional analysis of baseline data from prospective study	Russia	20–59	Community‐based population	427	Taking antihypertensive treatment or SBP ≥140 mm Hg or DBP≥90 mm Hg	NA

Abbreviations: BMI, body mass index; BP, blood pressure; CHD, coronary heart disease; DBP, diastolic blood pressure; NA, not available; SBP, systolic blood pressure.

3.2. Results of meta‐analysis

All eight studies demonstrated a positive association of hypertension status between spouses, although the ORs from two studies were not statistically significant.9, 12 The excess risk associated with spousal hypertension varied from as low as 15% (OR, 1.15; 95% confidence interval [CI], 1.06–1.25) in the ARIC (Atherosclerosis Risk in Communities Study) cohort by McAdams DeMarco and colleagues13 to approximately 120% (OR, 2.23; 95% CI, 1.75–2.72) in the case‐control study by Hippisley‐Cox and colleagues.7

By random‐effects model pooling ORs that were adjusted for most confounders in each study, the effect estimate for hypertension in patients whose spouses had hypertension was 1.41 (95% CI, 1.21–1.64) (Figure 2). Publication bias was suggestive by inspection of funnel plot (Figure S1); however, Egger test (P = .118) and Begg test (P = .536) did not indicate significant publication bias. A large degree of between‐study heterogeneity was observed (I ² = 80.2%). We then found that the case‐control study by Hippisley‐Cox and colleagues explained the majority of heterogeneity,7 with the I ² dropping to 15.1% (P = .315) when excluding this study.

Figure 2.

Forest plots showing odds ratios (ORs) and 95% confidence intervals (CIs) of spousal concordance for hypertension

Three studies reported results of linear correlations for BPs.7, 12, 13 Among those, one study showed stronger concordance of systolic BP than diastolic BP (r = .41 vs .25),7 whereas one study showed stronger concordance of diastolic BP than systolic BP (r = .19 vs .13),12 and another study suggested no difference.13 Meta‐analysis of the three studies did not suggest a significant difference between the concordance of systolic BP (r = .23, 95% CI, 0.02–0.43) and diastolic BP (r = .20, 95% CI, 0.11–0.27).

3.3. Sensitivity and subgroup analyses

Sensitivity analyses were performed by removing one study at a time and recalculating the pooled ORs with the remaining studies (Figure 3). We identified one outlier study that had major influence on the pooled results.7 Removing this study led to a pooled OR of 1.22 (95% CI, 1.14–1.31). No other study had significant influences on the results. Excluding the only study that did not adjust for age did not alter the associations (OR, 1.39; 95% CI, 1.18–1.66).14

Figure 3.

Sensitivity analyses by removing one study at a time and recalculating the pooled odds ratios (ORs) of the remaining studies

We also performed a series of secondary or subgroup analyses. First, four studies gave effect estimates that used wives’ hypertension statuses to predict the odds of hypertension in husbands and this generated a pooled OR of 1.64 (95% CI, 1.12–2.39).7, 10, 13, 14 Second, the pooled OR with BMI adjusted was 1.51 (95% CI, 1.08–2.12), similar to that from studies without BMI adjusted (OR, 1.48; 95% CI, 1.23–1.80) (Figure S2). Third, a greater magnitude of association was found among studies that defined outcome using diagnosed hypertension (OR, 1.53; 95% CI, 1.13–2.09), compared with studies that defined outcome using both diagnosed and undiagnosed hypertension (OR, 1.22; 95% CI, 1.08–1.39) (Figure S3).

We were also interested in the effect of cohabitation on the spousal concordance of hypertension. Such an effect, however, was scarcely quantitatively examined, because none of the studies provided OR values for the stratified groups that had different marriage lengths. We then conducted descriptive analyses here. Specifically, Jurj and colleagues11 reported a significantly stronger association for spouses who cohabited shorter than 23 years than spouses who cohabited longer, while Dolgalev and colleagues14 observed a stronger association for older couples. On the other hand, the length of marriage did not appear relevant in the studies by Konnov and colleagues12 and Stimpson and colleagues.10 Therefore, available studies did not provide conclusive evidence. It is possible that both assortative mating and cohabitation, either interactively or independently, contribute to the spousal concordance for hypertension.

4. DISCUSSION

4.1. Main findings

This is the first meta‐analysis that investigated the spousal concordance for hypertension. Our analyses demonstrated that spouses of individuals with hypertension had a 41% increased risk of having hypertension themselves. The association applied to both women and men, was independent of BMI, and was more pronounced in studies investigating diagnosed hypertension than studies investigating all hypertension (diagnosed and undiagnosed).

4.2. Spousal concordance for diseases

Marriage provides the primary source for social support and contexts for married people. It has been shown that marriage is associated with better physical and psychological well‐being, as well as low mortality rates.20, 21, 22 However, by sharing a common living environment, financial resources, and lifestyle habits, both partners can also share some health risk factors that may ultimately translate into diseases that they may otherwise not have in an alternative social environment. A number of previous studies have reported spousal similarities for mental health, diabetes mellitus, cardiovascular risk factors, and even cancers.23, 24, 25 A recent meta‐analysis showed that spousal history of diabetes mellitus conferred a 26% increased risk of diabetes mellitus.16 Another comprehensive meta‐analysis found that a variety of major coronary risk factors including smoking, BMI, and blood lipids were correlated between spouses.15 In line with these findings, our current meta‐analysis again suggests the existence of environmental causes during the development of metabolic and cardiovascular disorders.

4.3. Source of heterogeneity

There was moderately large heterogeneity among the included studies. Through forest plot and sensitivity analyses we identified one outlier study that accounted for the majority of heterogeneity: the case‐control study conducted by Hippisley‐Cox and colleagues showed large two‐ to three‐fold values of association between spouses.7 After excluding this study, the I ² was reduced to 15.1% and the overall OR point estimate was reduced to 1.22. The reason is not fully known, while possible explanations may include the case‐control design and the use of diagnosed hypertension in this study. This methodology was subject to recall bias and healthcare‐seeking bias. That is, spouses of patients with hypertension (cases) have a greater opportunity or initiative to access medical care than spouses of individuals without hypertension, which may lead to a greater likelihood of hypertension detection. This may only be resolved by including both diagnosed and undiagnosed cases as outcomes. Our secondary meta‐analysis by different ascertainments of hypertension also suggested that using diagnosed hypertension as an outcome variable may bias the effect estimate upward.

4.4. Risk of bias assessment

Four studies defined hypertension based on medical records or self‐report status.7, 8, 10, 11 These studies might have biased the effect estimates upward. Moreover, the studies by Konnov and colleagues and Dolgalev and colleagues only provided unadjusted ORs, which might also have biased the effect estimates upward.12, 14 The ARIC trial used repetitive BP measurements together with physician's diagnosis of hypertension across four visits spanning a period of 9 years of follow‐up and extensively adjusted for lifestyle and diet factors, which minimized the risk of bias.13

4.5. Assortative mating or cohabitation

Spousal concordance may operate through the effects of assortative mating or cohabitation. Assortative mating refers to the fact that people tend to choose a partner with similar characteristics, socioeconomic positions, and lifestyle patterns. If assortative mating is the main cause, then the observed association would be strongly reduced when adjusting for BMI (a surrogate for assortative mating). However, this is not the case. On the other hand, if the effect of cohabitation is the main cause, then the association would be strengthened along with the increasing length of marriage. However, the pattern was inconsistent across the four informative studies: in one study it was strengthened,14 in two studies it was unchanged10, 12 and in another it was reduced11 Therefore, the available study did not provide conclusive evidence. In fact, differential effects of assortative mating and cohabitation are not mutually exclusive. The two factors may operate jointly and separately to the spousal resemblance and should both be considered as correct interpretations for environmental contributors to hypertension.

4.6. Public and clinical significance

Together with the aging of society and changing of lifestyles, the number of cases of hypertension continues to grow. Although the current public strategy is based on wide screening of hypertension, there are still gaps in detection of undiagnosed hypertension, especially in developing countries where medical resources are limited and proper surveillance is deficient.26 The observation of concordance for couples in hypertension raises the possibility of using preexisting information of one partner to guide the screening of the other partner. A smart message conveyed to public health practitioners is that hypertension screening programs should be arranged for people whose spouses have hypertension.

Previous studies have found a “ripple effect” between spouses, suggesting that intervention programs targeted at individuals may also affect their spouses.27 For example, spouses in a dietary intervention program reduced their dietary fat and body weight to a larger degree than spouses in a control group.28 Therefore, interventions in hypertension could also be tailored to couples rather than individuals in order to make greater achievements. Previous hypertension prevention and intervention strategies focusing on individuals may shift to couple‐based interventions that enhance support and collaboration between spouses.

4.7. Study Limitations

First, all included studies except one applied a cross‐sectional design, and the time sequence of disease occurrence and marriage is not clear. Therefore, our study suggested a concordance rather than a causal relationship. Second, there was one study that did not adjust for age and the confounders adjusted in each study varied widely, which remained an inherent heterogeneity of included studies. Third, most included studies did not collect important lifestyle risk factors for hypertension such as exercise and diet, yet lifestyles were among the most similar factors between spouses. Therefore, it is not known whether the association of hypertension between spouses was mediated by those unmeasured factors. Fourth, publication bias might have existed although statistical values did not reach significance. This was probably because of the relatively small number of studies included in this meta‐analysis. Moreover, although our meta‐analysis demonstrated that the pooled ORs were not significantly different between studies that adjusted for BMI and those did not, the results should be interpreted with caution because there were only a small number of studies among each subgroup.

5. CONCLUSIONS

Our meta‐analysis demonstrated that spouses of individuals with hypertension were at an approximately 40% increased odds of having hypertension themselves. The concordance applied to both men and women and was unchanged after adjusting for BMI. The findings from our meta‐analysis suggest the importance of shared environmental factors during the development of hypertension. Spousal hypertension represents a robust signal for risk of hypertension and could facilitate its detection and screening. Future studies may need to adopt comprehensive study designs and include an extensive set of lifestyle variables in order to fully understand the mechanisms through which concordance manifests. Moreover, the cost and efficiency of screening and intervention programs that specifically target the spouses of patients with hypertension should be evaluated.

CONFLICT OF INTEREST

The authors have no conflicts of interest to declare.

Supporting information

 

Wang Z, Ji W, Song Y, et al. Spousal concordance for hypertension: A meta‐analysis of observational studies. J Clin Hypertens. 2017;19:1088–1095. 10.1111/jch.13084