Detecting undiagnosed hypertension using repeated blood pressure measurements: a cross-sectional study in rural Sidama Region, Ethiopia

Abstract

Objective

To determine the prevalence of undiagnosed hypertension and its risk factors among adults in rural Sidama Region, Ethiopia, using a two-step diagnostic method.

Design

A community-based cross-sectional study was conducted from 1 April to 31 July 2024. Data were collected among adults aged 45 years and above using the World Health Organization STEPwise Approach to Surveillance questionnaire. The Demographic and Health Survey questionnaire was also used to collect data on household characteristics.

Setting

Selected rural kebeles of Shebedino district, Sidama, Ethiopia.

Participants

2875 adults aged ≥45 years identified via census.

Outcome measures

Undiagnosed hypertension was defined as systolic blood pressure ≥140 mm Hg and/or diastolic blood pressure ≥90 mm Hg, in individuals with no history of the condition.

Results

The prevalence of undiagnosed hypertension ranged from 7.7% (95% CI: 6.7% to 8.7%) to 14.3% (95% CI: 13.0% to 15.6%). The previously diagnosed hypertensive cases were found in 3.3% (95% CI: 2.7% to 4.1%). Female sex (AOR 2.02; 95% CI: 1.44 to 2.82), age ≥ 65 years (AOR 1.48; 95%CI: 1.01 to 2.15), and history of alcohol drinking and khat chewing (AOR 2.94; 95%CI: 1.52 to 5.66) were significantly associated with undiagnosed hypertension. Lack of awareness of salt-related health risks (AOR 3.14; 95% CI: 2.30 to 4.30) and no prior blood pressure measurement (AOR 5.60; 95% CI: 1.73 to 18.07) were also associated with undiagnosed hypertension.

Conclusions

Undiagnosed hypertension is common among adults aged ≥45 years in the rural Sidama Region. Female sex, older age, substance use, limited awareness of salt-related health risks, and lack of prior blood pressure measurement were the identified risk factors. Regular screening should be implemented to detect cases at an early stage.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• The use of census data helped to ensure equitable access to screening for the entire eligible population.

• The two-step diagnostic method improved the reliability of the prevalence estimate.

• The use of sensitivity analysis to account for participant attrition minimizes selection bias and guarantees vigorous prevalence estimates despite the loss to follow-up.

• Behavioral factors like khat chewing, cigarette smoking, and alcohol drinking were self-reported, which may lead to social desirability bias.

Introduction

Hypertension is a chronic non-communicable disease, and it is the main risk factor for cardiovascular complications like heart attack, kidney failure, chest pain and stroke.¹ The prevalence of hypertension is increasing in low and middle-income countries.² In Ethiopia, around 27% of adults aged 30-79 are estimated to be hypertensive, and it is among the top causes of noncommunicable disease mortality in the country.^{3 4}

Hypertension is often asymptomatic, resulting in a high proportion of undiagnosed cases. Globally, about 45% of adults with hypertension are unaware they have the condition, and fewer maintain regular follow-up.² A study from rural central Ethiopia reported a prevalence of 21.3% undiagnosed hypertension among adults aged 19–65 years.⁵ Similarly, in the southern region of Ethiopia, 39.2% of adults aged 50 years or above were found to have undiagnosed hypertension.⁶

There are both non-modifiable and modifiable risk factors for hypertension. The non-modifiable risk factors include being older, having a family history of hypertension, sex and genetics. Modifiable risk factors include unhealthy eating habits (high salt intake and low fruit/vegetable consumption), physical inactivity, smoking, alcohol consumption and overweight/obesity.⁷

The prevalence of hypertension also varies by nutritional status across populations. A Nigerian study reported a higher hypertension risk among individuals with obesity compared with those who were underweight.⁸ Similarly, a study conducted among adults at the national level of Ethiopia found that underweight individuals had a lower risk of hypertension compared with those of normal weight.⁹

While several hypertension-related studies have been conducted in Ethiopia, most have focused on urban populations.10,13 However, rural communities face challenges that include limited healthcare access and lower health-seeking behaviour.^{14 15} These factors significantly contribute to the underdiagnosis of the disease.

To confirm the diagnosis of hypertension, blood pressure should be measured repeatedly to account for physiological variability. This, in turn, can reduce misclassification of the diagnosis.^{16 17} Despite this, most prior studies have relied on single-day blood pressure measurements to diagnosis the disease.1218,20 To fill this gap, our study employed a community-based screening followed by confirmation at a health centre.

Therefore, the objective of the study was to determine the prevalence of undiagnosed hypertension and its risk factors among adults in rural Sidama Region, Ethiopia, using a two-step diagnostic method.

Materials and methods

Study design

A community-based cross-sectional study was conducted, and we adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) checklist (online supplemental table 1).

Study setting and period

The study was conducted in the Sidama National Regional State of Ethiopia from 1 April to 31 July 2024. Sidama Region is known for its fertile highlands and has an estimated total population of around five million.²¹ The regional capital, Hawassa, is located about 275 kilometres from Addis Ababa and serves as the main administrative, commercial and health service centre of the region.²²

Shebedino District, one of the 36 administrative districts in Sidama Region, was purposively selected for this study. It is situated about 310 kilometres from Addis Ababa and has an estimated population of 326 871 (164 707 males and 1 62 164 females).^{21 23} The district comprises 23 rural and 9 urban kebeles (the smallest administrative units) and is served by one general hospital and six health centres supported by satellite health posts.

Dobetoga is one of the rural catchment areas in Shebedino district, selected as the study site for its broad rural coverage. The catchment comprises four kebeles: Dobetoga, Gobe Hebisha, Gonowa Gabalo and Howolso, all of which were included in this study (figure 1). Dobetoga Health Centre serves the population of the catchment. The health centre offers comprehensive maternal and child health, emergency and diagnostic services to its catchment population.

Figure 1. Map of the study area.

Population

Adults aged 45 years and older who had resided in the selected kebeles for at least 6 months were included in the study. Pregnant women and critically ill individuals were predefined as exclusion criteria to control their confounding effects on blood pressure readings.

During the data collection time, no pregnant participants were identified, likely because eligibility was restricted to adults aged 45 years and above. Furthermore, although the proposal accounted for critical illness, a condition often expected at older ages, no such cases were encountered among the participants. This may be because people with severe illnesses were already admitted to health facilities.

Study variables and their measurements

Different sociodemographic characteristics of the study population were assessed in this study. It includes sex, marital status, educational level, and occupational status. Household-level wealth index and distance from the nearby health facility were also assessed. Behavioral characteristics were also interviewed, including cigarette smoking, khat chewing, and alcohol consumption. Dietary behaviors like fruit and vegetable intake and Salt consumption related questions were also asked. Their physical activity situations were also determined, including the intensity and its frequency. We also took anthropometric measures (height, weight, waist circumference, body mass index, and waist-to-height ratio).

The Demographic and Health Survey tool was used to determine household wealth status.²⁴ It was calculated based on a composite score of ownership of household assets and characteristics. We applied the Principal Components Analysis principle to categorize the wealth status by transforming correlated asset variables into a single wealth score. Variables that do not have variation and those with extreme values (asset scores below the 5^th or above the 95^th percentile) were excluded to minimize the effect of outlier.²⁵

The anthropometric measurements were taken using the standard procedures. All devices were calibrated before measurements. Each team measured 15 individuals twice to ensure measurement reproducibility and validity. To measure the height, the person stands upright with his/her back straight and feet together. The measuring instrument was properly calibrated. Measurement was taken from the top of the head to the floor, without shoes. While measuring weight, the person stands on a flat and level surface. The scale was properly calibrated and located on a solid and stable surface. They were dressed in lightweight clothing and wore no shoes. The scale was reset to zero before each measurement. The weight measurements were taken to the nearest 0.1 kg.

To measure waist circumference, the tape measure was placed around the body at the top of the hip bone, usually at the level of the belly button. The tape measure was snug but not too tight and was parallel to the floor. To measure hip circumference, the individual stood upright with his/her feet together and weight evenly distributed. A flexible, non-stretchable tape was placed horizontally around the widest part of the buttocks, typically between the greater trochanter and the lower buttock level. The tape rests snugly against the skin without indenting it. The measurements were taken to the nearest 0.1 cm.²⁶

Body mass index, which is a proxy for body fat, was calculated as weight (kg)/height (m²). We categorised based on the World Health Organization guidelines. Underweight (<18.5 kg/m²), normal (18.5–24.9 kg/m²), overweight (25.0–29.9 kg/m²) and obese (≥30.0 kg/m²).²⁶

The outcome variable is undiagnosed hypertension. It was defined as a newly diagnosed hypertensive case among the surveyed population. New hypertension cases were defined as participants with no prior diagnosis and whose systolic blood pressure was ≥140 mm Hg and/or diastolic blood pressure ≥90 mm Hg across both community and facility level measurements.²⁶ The history of hypertension diagnosis was determined by asking them if a healthcare provider had ever informed them that they are hypertensive or provided them with related treatment or advice.¹⁶

Diagnostic criteria for hypertension

Blood pressure was measured following the standard procedure. Participants were instructed to avoid caffeine and any vigorous-intensity activity or exercise for at least 30 minutes before the measurement. During measurement, they were seated with their back support, feet flat on the floor, and legs uncrossed. Any cloth that covered the arm was removed, and both the participant and healthcare provider abstained from speaking during the measurement. These steps were followed based on recommended practices and guidelines.^{16 17 27} The data collectors followed standard device procedures: cuff size matched upper-arm circumference, arm rested on a flat surface and the device automatically recorded systolic/diastolic blood pressure and pulse rate.

Three blood pressure readings were taken in the community, and each reading was documented. The individuals rested for 15 min before the first measurement. After a 5-minute interval, the second reading was taken. In a similar way, the person waited for another 5 min, and the third measurement was carried out. The average of the three readings was used to decide on the blood pressure level. A calibrated aneroid sphygmomanometer and stethoscopes were used to measure blood pressure and recorded to the nearest 2 mm Hg.

Based on community-based screening, individuals whose average systolic blood pressure was ≥140 mmHg and/or diastolic blood pressure ≥ 90 mmHg were identified and referred to the health center for diagnostic confirmation. Referred individuals were told to attend their referral after one week (seven days), in line with the National Institute for Health and Care Excellence guidelines.²⁸ If the seventh day fell on a weekend, the visit was postponed to the next working day. Upon arrival at the health centre, attendees were given a minimum of 30 min rest before their blood pressure was re-measured using the same standardised procedure, and the diagnosis was then confirmed.

Census procedure

All permanent residents of the study sites who met the target (age 45 years and above) were included in the study. A census was conducted instead of a sample to include all eligible individuals and to increase the statistical power. The Inclusion of all eligible residents also helped us to ensure adherence to the ethical principle of including all eligible community members. To maximize the response rate, data collectors used a list of households obtained from the local administrative unit (Kebeles) to identify every eligible resident. Accordingly, 2,875 individuals were identified and screened.

Assessment of outcomes through two-stage measurements

To diagnose hypertension, two steps were followed. In the first step, blood pressure screening was conducted by trained health professionals in the community. A total of 2875 adults were screened across four kebeles: Dobetoga (n=653), Gobe Hebisha (n=804), Gonowa Gabalo (n=725) and Howolso (n=693). Of those screened, 625 met the threshold limit for being suspected of hypertension (blood pressure (BP) ≥140 mm Hg and/or diastolic BP ≥90 mm Hg). Among these, 529 were newly identified, whereas 96 had a prior diagnosis of hypertension.

In the second step, individuals suspected of having hypertension were referred to the nearest health centre for final confirmation. Among the newly identified suspected cases, 339 (64.1%) attended their follow-up, of whom 221 (65.2%) were confirmed to have hypertension. Additionally, 47 of the 96 previously diagnosed cases attended their follow-up, with 30 of them confirmed as hypertensive (figure 2).

Figure 2. Flow diagram of hypertension (HTN) case ascertainment across study sites.

Abbreviations: PHCU, Primary Health Care Unit; HC, Health Center.

Data Collection Tools and Procedures

We used the World Health Organization STEPwise approach to surveillance for non-communicable diseases questionnaire for the data collection.²⁶ The Demographic and Health Survey household questionnaire was additionally used to collect data on household characteristics. The tool was initially prepared in English and translated into the local language (Sidaamu Affoo) by bilingual experts, then back translated into English by independent bilingual experts to ensure consistency.

Data were collected using physical measurements (blood pressure and anthropometry) and face-to-face interviewer-administered questionnaires. Eight trained health care providers (two per team) served as data collectors under close supervision. They were selected based on their qualifications, experience, and fluency in the local language. We provided a three-day training course for both data collectors and supervisors. A pretest took place with 5% of the total study participants in Remeda kebele, in Hawella district, located outside the study area. The European server of Kobo Toolbox (KoBoToolbox, n.d.) was used for electronic data capture in the field.

Supervisors were providing technical support during the data collection time. A deadline was set for data submission, and we were providing regular feedback through telegram, phone calls, and face-to-face contact. The data was secured, and Specific access privileges were given to each data collector. They were collecting the data, saving files in the drafts, and submitting them to the server. But they were not downloading or editing any data.

Data management and analysis

The collected data were exported to Stata release 17 (StataCorp, 2021, College Station, TX: StataCorp LLC) for analysis. Completeness, internal consistency and potential outliers in quantitative variables were examined and corrected as necessary. Derived variables, such as body mass index and waist-to-height ratio, were then computed. Normality of continuous variables was assessed using visual methods (histograms with normal curves) and statistical measures of skewness and kurtosis.

After the data were cleaned, descriptive analyses were performed to characterise the study population. Independent-samples t tests were used to compare mean anthropometric measures between males and females after assumptions of normality and linearity were verified. Pearson correlation analyses were done to examine associations between anthropometric measures and systolic and diastolic blood pressure.

Subgroup analyses was performed to compare baseline sociodemographic ch aracteristics between participants who attended their referral and those who did not. Categorical variables were compared between health centre attendees and non-attendees using Pearson’s chi-square test. These variables included sex, marital status, occupational status, literacy status, time to reach the health facility, wealth index, and community-based health insurance membership and renewal status. The Wilcoxon rank-sum (Mann–Whitney U) test was used to compare participants’ age by health centre attendance status. The tabstat command was used to estimate the median age with its interquartile ranges.

A bivariate logistic regression analysis was performed to identify candidate variables for multivariable model. Variables with p≤0.25 in bivariate analysis were considered candidates for multivariable modelling to minimise the risk of excluding potentially important covariates. In addition, variables identified a priori based on epidemiological theory and prior evidence were included in the multivariable model regardless of their statistical significance.

Three analytic models: standard logistic regression, mixed-effects and generalised estimating equations were applied to estimate risk factors, with villages served as clustering units. Estimates from the logistic regression and generalised estimation equation models were almost similar, but the mixed-effects model showed a slight difference. An intraclass correlation coefficient of 0.6% with the mixed model suggested minimal variation between clusters. Although the mixed-effects model was statistically valid, the low level of clustering suggested limited additional benefits over marginal models. Then, a comparison between the standard logistic regression and the generalised estimating equation model followed. The logistic regression model had an Akaike Information Criterion (AIC) of 1476.2 and an area under the Receiver Operating Characteristic (ROC) curve of 0.71. The generalised estimating equation model showed a Quasi Information Criterion (QIC) of 1628.5, with the same area under the ROC curve of 0.71. While the standard logistic regression model had a lower information criterion, it lacks a mechanism for within-cluster correlations. This absence risks biased standard errors and unreliable population inferences. Therefore, the generalised estimating equation model was the best choice for the final interpretation. Its design ensures robustness to intracluster correlation and suitability for population-averaged estimates.²⁹ Online supplemental table 2 contains additional details.

Both confounding and interaction effects were assessed in the final model. Potential confounding effect was assessed by examining changes in regression coefficients following the sequential inclusion of covariates. The regression coefficient for alcohol consumption shifted by more than 10% after khat chewing was added, which indicates a potential confounding effect. The model, therefore, included an interaction term between alcohol consumption and khat chewing. The interaction term was statistically significant, indicating the presence of effect modification, and was retained in the final model. Results are presented as adjusted odds ratios (AOR) with 95% confidence intervals (CI).

Reliability of measurements

To assess the consistency of repeated blood pressure measurements, we calculated intraclass correlation coefficients for the three community-based readings. Furthermore, the agreement between community-based screening and facility-level confirmation was evaluated using the Bland-Altman method.³⁰ It quantifies the mean difference and limits of agreement between the two settings.

Prevalence estimation and sensitivity analysis

The prevalence of undiagnosed hypertension was first calculated based on confirmed cases at the health centre. The subgroup analysis indicates that there were no significant differences in the basic sociodemographic characteristics between health centre attendees and non-attendees (see online supplemental table 3). Therefore, an adjusted prevalence was estimated by applying the observed confirmation rate of attendees to the non-attendee group.

To account for the uncertainty of this estimate, a sensitivity analysis was performed using two extreme scenarios. The first scenario assumes all non-attendees were normotensive, and the second scenario assumes all non-attendees were hypertensive.

Ethical clearance

This study was reviewed and approved by Hawassa University College of Medicine and Health Science Institutional Review Board. An ethical clearance letter was provided (Ref. No: IRB/101/16; 12/03/2024). We also obtained an official support letter from the Sidama National Regional State Public Health Institute.

Informed consent was obtained from all participants via signature or thumbprint before data collection. For illiterate respondents, the consent form and study objectives were read aloud. To protect participant dignity and encourage candid responses, all interviews were conducted in private settings to ensure strict confidentiality.

Result

Sociodemographic characteristics of adults in the rural district of Sidama Region, Ethiopia

In this study, 2,875 adults participated, and 1,586 (55.2%) of them were female. Nearly half of the participants, 1398 (48.6%), were aged 45–54 years, and 2553 (88.8%) were currently married. Regarding occupation, 1353 (47.0%) were farmers, and 1482 (51.6%) were housewives. Most participants, 2513 (87.4%), were unable to read or write, and 2575 (89.6%) had never received formal education.

Most participants, 2370 (82.4%), took 30 min or more to reach the nearest health facility. Regarding wealth status, 575 (20.0%) were in the poorest quintile, while 502 (17.5%) were in the richest. Only 990 (34.4%) were members of the Community-Based Health Insurance scheme, and among these, 815 (82.3%) had not renewed their membership in that year (table 1).

Table 1. Sociodemographic characteristics of adults in the rural district of Sidama Region, Ethiopia (n=2875).

Variables	Categories	Existing HTN N (%)	Undiagnosed HTN N (%)	Screened+, not confirmed n (%)	Non-HTN N (%)	Total N (%)
Sex	Male	42 (43.8)	83 (37.6)	77 (40.5)	1087 (45.9)	1289 (44.8)
	Female	54 (56.2)	138 (62.4)	113 (59.5)	1281 (54.1)	1586 (55.2)
Age	45–54	28 (29.2)	112 (50.7)	73 (38.4)	1185 (50.0)	1398 (48.6)
	55–64	41 (42.7)	59 (26.7)	63 (33.2)	747 (31.6)	910 (31.7)
	65+	27 (28.1)	50 (22.6)	54 (28.4)	436 (18.4)	567 (19.7)
Marital status	Not married	11 (11.5)	29 (13.1)	32 (16.8)	250 (10.6)	322 (11.2)
	Married	85 (88.5)	192 (86.9)	158 (83.2)	2118 (89.4)	2553 (88.8)
Occupation	Farmer	49 (51.0)	88 (39.8)	79 (41.6)	1137 (48.0)	1353 (47.0)
	Housewife	43 (44.8)	131 (59.3)	107 (56.3)	1201 (50.7)	1482 (51.6)
	Other	4 (4.2)	2 (0.9)	4 (2.1)	30 (1.3)	40 (1.4)
Literacy status	Read/write	18 (18.8)	22 (9.9)	27 (14.2)	295 (12.5)	362 (12.6)
	Illiterate	78 (81.2)	199 (90.1)	163 (85.8)	2073 (87.5)	2513 (87.4)
Educational status	Formal	13 (13.5)	17 (7.7)	28 (14.7)	242 (10.2)	300 (10.4)
	None	83 (86.5)	204 (92.3)	162 (85.3)	2126 (89.8)	2575 (89.6)
Travel time at HF	<30 min	22 (22.9)	49 (22.2)	42 (22.1)	392 (16.6)	505 (17.6)
	≥30 min	74 (77.1)	172 (77.8)	148 (77.9)	1976 (83.4)	2370 (82.4)
Wealth index	Poorest	12 (12.5)	39 (17.7)	37 (19.5)	487 (20.6)	575 (20.0)
	Poorer	10 (10.4)	37 (16.7)	33 (17.3)	520 (21.9)	600 (20.9)
	Middle	18 (18.8)	56 (25.3)	45 (23.7)	431 (18.2)	550 (19.1)
	Richer	18 (18.8)	53 (24.0)	37 (19.5)	540 (22.8)	648 (22.5)
	Richest	38 (39.5)	36 (16.3)	38 (20.0)	390 (16.5)	502 (17.5)
CBHI membership	Member	42 (43.8)	58 (26.2)	61 (32.1)	829 (35.0)	990 (34.4)
	Non-member	54 (56.2)	163 (73.8)	129 (67.9)	1539 (65.0)	1885 (65.6)
CBHI renewal (n=990)	Updated	8 (19.1)	15 (25.9)	14 (22.9)	138 (16.7)	175 (17.7)
	Not updated	34 (80.9)	43 (74.1)	47 (77.1)	691 (83.3)	815 (82.3)
Total		96 (100)	221 (100)	190 (100)	2368 (100)	2875 (100)

CBHI, community-based health insurance; HF, health facility; HTN, hypertension.

Substance use, dietary habits and physical activity status

Most study participants reported no history of smoking (2583; 89.8%), alcohol use (2338; 81.3%) or khat chewing (2651; 92.2%). Exposure through family was also minimal. Regarding diet, nearly half of the respondents eat fruit between meals, and vegetable consumption was also common. Salt intake perception was a concern: 2013 (70.1%) believed they were consuming the right amount of salt, while 1725 (60.0%) were aware that too much salt could harm their health.

Drinking coffee is common among the study population. A total of 2793 (97.2%) individuals reported that they drink coffee. Among them, 1742 (62.4%) consumed 3 to 5 cups per day, and 783 (28.0%) consumed 6 or more cups per day. Regarding physical activity, 1340 (46.6%) of respondents engaged in vigorous tasks, and 749 (26.1%) participated in moderate intensity activities (table 2).

Table 2. Behavioural and lifestyle characteristics of study participants in the rural districts of Sidama Region, Ethiopia (n=2875).

Variables	Category	Frequency (n)	Percent (%)
Cigarette smoking history	Yes	292	10.2
	No	2583	89.8
Family members smoke cigarettes	Yes	50	1.7
	No	2825	98.3
Friends smoke cigarettes	Yes	93	3.2
	No	2782	96.8
Alcohol drinking history	Yes	537	18.7
	No	2338	81.3
Family members drink alcohol	Yes	136	4.7
	No	2739	95.3
Friends drink alcohol	Yes	136	4.7
	No	2739	95.3
Khat chewing history	Yes	224	7.8
	No	2651	92.2
Eating fruit between meals	Yes	1333	46.4
	No	1542	53.6
Eating vegetables between meals	Yes	1843	64.1
	No	1032	35.9
Thought on the amount of salt consumed	Too much	50	1.7
	Just the right amount	2013	70.1
	Too little	812	28.2
Thought on the consumption of too much salt causes health problems	Yes, it causes	1725	60.0
	No, it does not cause	838	29.2
	Don’t know	312	10.8
Currently drink coffee	Yes	2793	97.2
	No	82	2.8
Number of cups of coffee consumed per day	1–2	268	9.6
	3–5	1742	62.4
	6 or more	783	28.0
Add salt while drinking coffee	Yes	2666	95.5
	No	127	4.5
Work vigorous intensity activity	Yes	1340	46.6
	No	1535	53.4
Work moderate intensity activity	No	2126	73.9
	Yes	749	26.1

Anthropometric assessment

Anthropometric measures were significantly higher in males than females. Mean waist circumference was 69.9 (SD 8.5) cm in males and 68.1 (SD 7.9) cm in females (p<0.001). The mean body weight was 50.1 (SD 7.5) kg among males compared with 44.5 (SD 6.4) kg among females (p<0.001). Mean body mass index was 19.2 (SD 3.0) kg/m² among males and 18.7 (SD 2.7) kg/m² among females (p<0.001). In contrast, the waist-to-height ratio was slightly higher in females than males (44.2 (SD 5.4) vs 43.2 (SD 5.4); p<0.001).

Waist circumference showed positive correlations with systolic and diastolic blood pressure (r=0.16 and r=0.17, respectively; both p<0.001). Similar positive correlations were observed between waist-to-height ratio and systolic (r=0.15) and diastolic (r=0.17) blood pressure, with both associations statistically significant (p<0.001). Body mass index was also positively correlated with systolic (r=0.15) and diastolic (r=0.17) blood pressure, with both correlations being significant (p<0.001) (table 3).

Table 3. Anthropometric measures and their correlation with systolic and diastolic blood pressure.

Anthropometric measures	Mean (SD)				Correlation with systolic blood pressure	Correlation with diastolic blood pressure
	Male	Female	All	P-value	r (P-value)	r (P-value)
Waist circumference(cm)	69.9 (8.5)	68.1 (7.9)	68.9 (8.3)	<0.001	0.16 (<0.001)	0.17 (<0.001)
Waist-to-height ratio	43.2 (5.4)	44.2 (5.4)	43.8 (5.4)	<0.001	0.15 (<0.001)	0.17 (<0.001)
Weight(kg)	50.1 (7.5)	44.5 (6.4)	47.0 (7.5)	<0.001	0.16 (<0.001)	0.16 (<0.001)
Body mass index(kg/m2)	19.2 (3.0)	18.7 (2.7)	18.9 (2.9)	<0.001	0.15 (<0.001)	0.17 (<0.001)

Reliability of the repeated blood pressure measurements

The mean systolic blood pressure declined from 126.8 to 122.7 mm Hg across the three community-based readings, while the mean diastolic pressure decreased from 84.2 to 82.3 mm Hg. The pooled mean of the three readings was 124.4 (SD 21.5) mm Hg for systolic blood pressure and 83.1 (SD 12.7) mm Hg for diastolic blood pressure. The reliability of the repeated measurements was high, with intraclass correlation coefficients of 0.96 (95% CI: 0.95 to 0.97; p<0.001) for systolic blood pressure and 0.95 (95% CI: 0.94 to 0.96) for diastolic blood pressure (see online supplemental table 4).

Based on the Bland-Altman analysis, the mean difference between community and health-centre level measurements was -3.9 mm Hg (95% limits of agreement −38.2 to 30.4) for systolic blood pressure. For diastolic blood pressure, the mean difference was −9.4 mm Hg (95% limits of agreement−28.5 to 9.7), reflecting lower average values at the health-centre level. These indicate that health-centre readings were lower on average, though the wide limits of agreement highlight significant individual variability between settings. The detail is presented in online supplemental table 5.

Homogeneity of the repeated blood pressure measurement across age and sex

Across the three repeated community-level measurements, systolic and diastolic blood pressure showed a slight decline over time in all age groups and by sex. For instance, in the first measurement, mean systolic blood pressure was 126.8 (SD 20.9) mm Hg in participants aged 45–54 years and 128.9 (SD 24.7) mm Hg in those aged 65 years or above. By the third measurement, these values had decreased modestly to 123.2 (SD 19.3) mm Hg and 123.6 (SD 23.8) mm Hg, respectively. Diastolic blood pressure showed a similar decline over the three measurements. In the first measurement, the mean diastolic blood pressure was 85.0 (SD 12.3) mm Hg among participants aged 45–54 years and 83.5 (SD 13.8) mm Hg in those aged 65 years or above. This has decreased to 83.5 (SD 12.0) mm Hg among individuals aged 45–54 years and 80.8 (SD 13.4) mm Hg among those aged 65 years or above by the third measurement. The trend was similar for both males and females (online supplemental table 6).

The intraclass correlation coefficients showed high within-individual consistency for both systolic and diastolic blood pressure. For systolic blood pressure, intraclass correlation coefficient ranged from 0.878 (95% CI: 0.868 to 0.888) to 0.888 (95% CI: 0.872 to 0.902) across age groups and from 0.881 (95%: CI 0.871 to 0.890) to 0.885 (95% CI: 0.874 to 0.894) by sex. For diastolic blood pressure, it ranged from 0.864 (95% CI: 0.853 to 0.875) to 0.889 (95% CI: 0.874 to 0.903) across age groups and from 0.859 (95% CI: 0.847 to 0.871) to 0.878 (95% CI: 0.868 to 0.887) by sex (see online supplemental table 7).

Major finding

The adjusted prevalence of undiagnosed hypertension was 12.0% (95% CI: 10.8 to 13.2). This estimate includes the 7.7% (221/2875; 95% CI: 6.8 to 8.7) confirmed cases at the health centre plus an estimated 4.3% from screen-positive individuals who did not attend follow-up. The 4.3% was derived by multiplying the number of non-attendees (190) by the 65.2% confirmation rate observed among those who did attend and then dividing by the total surveyed population (2875). This calculation assumed a similar hypertension risk between both groups and was justified by their comparable sociodemographic profiles (online supplemental table 6).

Sensitivity analyses indicated that the prevalence of undiagnosed hypertension remained at 7.7% (95% CI: 6.8 to 8.7) under the conservative assumption that all screen-positive but health centre non-attendees were not hypertensive. If all non-attendees were assumed to be hypertensive, the prevalence reached 14.3% (95% CI: 13.0 to 15.6).

In the community level survey, 3.3% (95%: CI 2.7 to 4.1) reported a prior diagnosis of hypertension. From these, 38.5% (95% CI: 29.3 to 48.7) were in regular follow-up, and 20.8% (95% CI: 13.8 to 30.2) were taking antihypertensive medication (table 4).

Table 4. Community-based hypertension screening, referral adherence and adjusted prevalence estimates (n=2875).

Category	Denominator (N)	Cases (n)	Prevalence (95% CI)
Total suspected hypertension cases in the community	2875	625	21.7 (20.3 to 23.3)
Previously diagnosed cases	2875	96	3.3 (2.7 to 4.1)
Cases on regular follow-up	96	37	38.5 (29.3 to 48.7)
Cases on antihypertensive medication	96	20	20.8 (13.8 to 30.2)
Newly identified cases (community screening)	2779	529	18.4 (17.0 to 19.9)
Attended a referral at the health centre	529	339	64.1 (59.9 to 68.1)
Confirmed hypertensive cases at the health centre	339	221	65.2 (59.9 to 70.1)
Minimum prevalence (Scenario 1)	2875	221	7.7 (6.8 to 8.7)
Adjusted prevalence	2875	345	12.0 (10.8 to 13.2)
Maximum prevalence (Scenario 2)	2875	411	14.3 (13.0 to 15.6)

Minimum prevalence (Scenario 1) assumes only cases confirmed at the health center are true hypertensive cases, whereas maximum prevalence (Scenario 2) assumes all suspected and confirmed cases are true hypertensive cases.

Crude, sex-stratified and combined multivariable regression analysis of risk factors for undiagnosed hypertension in rural Sidama Region, Ethiopia

The crude analysis identified sex, history of cigarette smoking, alcohol drinking, khat chewing, perceived salt intake problems, screening history for hypertension, and waist-to-height ratio as potential risk factors. In the adjusted model, female sex (AOR 2.02; 95% CI: 1.45 to 2.82) and age ≥ 65 years (AOR 1.48; 95% CI: 1.01 to 2.15) were independently associated with undiagnosed hypertension. The combined history of alcohol drinking and khat chewing had a higher risk (AOR 2.94; 95% CI: 1.52 to 5.66) than neither behavior. Lack of perceived salt intake problems (AOR 3.14; 95% CI: 2.30 to 4.30) and never having had a prior blood pressure measurement (AOR 5.60; 95% CI: 1.73 to 18.07) were the other significantly associated factors.

The global sex-by-covariate interaction test was marginally significant (χ² (13) =22.15; p=0.053), which suggests the presence of possible sex differences in the associations. Alcohol consumption was modified by khat chewing status among men (interaction AOR 4.38; 95% CI: 2.09 to 9.18), but no such modification was observed among women. Lack of moderate-intensity physical activity was associated with undiagnosed hypertension, specifically in women (AOR 1.72; 95% CI: 1.09 to 2.69), with no significant association in men (AOR 0.73; 95% CI: 0.43 to 1.24). Not perceiving a salt intake problem was a significant factor for both sexes, though the association was stronger in women (AOR 4.21; 95% CI: 2.81 to 6.31) than in men (AOR 2.39; 95% CI: 1.43 to 4.00). Similarly, never having a prior blood pressure measurement showed a greater association in women (AOR 8.52; 95% CI: 1.15 to 63.17) compared to men (AOR 4.66; 95% CI: 1.09 to 19.89) (table 5).

Table 5. Sex-stratified and combined multivariable regression analysis of risk factors for undiagnosed hypertension in rural Sidama Region, Ethiopia.

Variable	Category	Crude OR (95% CI)	Men AOR (95% CI)	Women AOR (95% CI)	Combined AOR (95% CI)
Sex	Male	1.00	NA	NA	1.00
	Female	1.40 (1.10 to 1.80)	NA	NA	2.02 (1.45 to 2.82)
Age	45 to 54	1.00	1.00	1.00	1.00
	55 to 64	0.80 (0.60 to 1.10)	0.68 (0.38 to 1.24)	1.09 (0.72 to 1.66)	0.91 (0.64 to 1.30)
	65+	1.10 (0.80 to 1.60)	1.75 (0.99 to 3.07)	1.18 (0.68 to 2.05)	1.48 (1.01 to 2.15)
Literacy status	Can read and write	1.00	1.00	1.00	1.00
	Cannot read and write	1.30 (0.80 to 2.10)	0.91 (0.49 to 1.67)	1.59 (0.66 to 3.81)	1.11 (0.68 to 1.80)
Household asset score	Per unit increase	1.03 (0.90 to 1.10)	0.96 (0.86 to 1.08)	1.08 (0.99 to 1.17)	1.04 (0.97 to 1.11)
History of cigarette smoking	No	1.00	1.00	1.00	1.00
	Yes	1.50 (1.01 to 2.20)	0.76 (0.41 to 1.44)	1.78 (0.76 to 4.19)	1.25 (0.75 to 2.06)
History of alcohol* drinking×khat chewingᵃ	No×no	1.00	1.00	1.00	1.00
	No×yes	2.00 (1.10 to 3.80)	2.93 (1.04 to 8.21)	2.82 (1.18 to 6.74)	2.81 (1.46 to 5.40)
	Yes×no	1.60 (1.10 to 2.20)	2.48 (1.34 to 4.59)	1.03 (0.54 to 1.97)	1.60 (1.04 to 2.45)
	Yes×yes	2.10 (1.20 to 3.60)	4.38 (2.09 to 9.18)	1.00 (1.00 to 1.00)	2.94 (1.52 to 5.66)
Perceived salt intake problem	Yes	1.00	1.00	1.00	1.00
	No	3.20 (2.40 to 4.30)	2.39 (1.43 to 4.00)	4.21 (2.81 to 6.31)	3.14 (2.30 to 4.30)
	Do not know	1.70 (1.10 to 2.70)	1.10 (0.49 to 2.45)	1.87 (0.99 to 3.50)	1.38 (0.84 to 2.25)
Moderate intensity physical activity	Yes	1.00	1.00	1.00	1.00
	No	1.10 (0.80 to 1.40)	0.73 (0.43 to 1.24)	1.72 (1.09 to 2.69)	1.30 (0.91 to 1.80)
Ever had BP measured	Yes	1.00	1.00	1.00	1.00
	No	5.70 (1.80 to 18.00)	4.66 (1.09 to 19.89)	8.52 (1.15 to 63.17)	5.60 (1.73 to 18.07)
Body mass index (mean (SD))	Per unit increase	1.04 (0.99 to 1.09)	1.05 (0.96 to 1.15)	1.00 (0.93 to 1.08)	1.02 (0.96 to 1.08)
Waist-to-height ratio (mean (SD))	Per unit increase	1.04 (1.02 to 1.07)	1.01 (0.96 to 1.07)	1.02 (0.98 to 1.06)	1.02 (0.99 to 1.05)

Combined AORs are adjusted for all covariates listed in the table

^*ᵃ Interaction term between history of alcohol drinking and khat chewing (reference category: no alcohol, no khat)

AOR, adjusted OR; BP, blood pressure; NA, not applicable.

Discussion

This study provided important evidence on the prevalence and contributing factors of undiagnosed hypertension in the rural community. While the overall prevalence was relatively low compared with some previous reports, a notable proportion of adults remained unaware of their condition.

According to this study, around 35% of adults whose blood pressure was ≥ 190/140 mmHg in the community had a normal readings when they attended their referral at the health centre. This result suggests the likelihood of masked hypertension (elevated out-of-clinic but normal clinic blood pressure reading).³¹ The study did not address white-coat hypertension, because screen-negative individuals were not referred to the health centre.³² From this result, we can understand that while community-level screening is important for identifying potential cases, it requires secondary confirmation to ensure accurate diagnosis.

This study found a lower prevalence of undiagnosed hypertension than previous research. Earlier rural studies in Central Ethiopia and Rwanda reported a prevalence of 21.3%⁵ and 18.7% ³³ respectively. A study from rural Uganda reported that the prevalence of hypertension was 27.3%, and 61.5% of cases were undiagnosed.³⁴ Another study from rural Nepal also reported that among patients with hypertension, 50.4% remained undiagnosed.³⁵

The lower prevalence in the current study may be related to the methodology. The repeated measurement might have reduced overestimation due to the temporary elevation of blood pressure.³⁶ A national health and nutrition examination survey reported that a single blood pressure measurement overestimated the prevalence of hypertension by 12.6%.³⁷ From this, we can understand the importance of repeated measurements to ensure diagnostic accuracy and avoid overestimation.

Although the prevalence was relatively low, many rural residents are living with undiagnosed hypertension. Several factors contribute to this gap. Limited education and low health literacy reduce awareness on chronic disease risks.³⁸ As a result, many people do not recognise hypertension as a threat. Access to healthcare services is the other factor. Long travel distances, financial barriers and low healthcare service utilisation affect the routine blood pressure check-up in rural areas.^{5 14} In addition, community-based screening programmes are limited in rural settings.³⁹ This leaves many adults without the opportunity to detect the cases at an early stage. The composite effect of these factors allows a significant portion of hypertension cases in rural communities to remain undiagnosed.

Different demographic, behavioural and lifestyle-related factors are associated with the likelihood of undiagnosed hypertension. The current study showed that women are more likely than men to have undiagnosed hypertension. This is in line with other studies in Bangladesh⁴⁰ and in India.⁴¹ A study in rural northwest Ethiopia also found that higher rates of hypertension are linked to female gender.⁴² On the other hand, a study from rural South Africa found no significant gender differences in hypertension among those over 55 years of age.⁴³ There have been other studies that found the reverse, with men having higher rates of undiagnosed hypertension than women. For instance, a survey from Mizan Aman and Gonder City in Ethiopia reported that men were more likely to have hypertension than women.^{13 44} Those studies included participants aged 18 years and above, and both were conducted in urban contexts. Thus, the contradictory findings may be attributed to differences in study settings and the state of healthcare services in the respective areas. The age range of study participants could also be another reason. The higher risk among women in the current study might be related to biological reasons. Women, after menopause, may face a greater risk because of hormonal changes that can raise blood pressure.⁴⁵ Lifestyle factors, such as higher salt intake among rural women, may also have contributed to the higher prevalence of the disease in women compared with men.⁴⁶

Adults aged 65 years and older have a higher odd of undiagnosed hypertension compared to those aged 45–54 years in this study. This aligns with other studies that found a positive association between old age and hypertension.1347,49 This is justifiable, especially in rural areas where older people often challenged to get healthcare access and do not have regular health check-ups.¹⁴ A study from the rural district of Bangladesh reported the occurrence of higher undiagnosed hypertension among elderly people due to low health-seeking behavior.⁵⁰

The odds of undiagnosed hypertension were higher among people who drink alcohol and chew khat at the same time. This is supported by findings that showed the synergistic effects of these substances' use on chronic diseases like hypertension.^{19 51} Other studies from Ethiopia also showed that individuals engaging in multiple risky behaviors, like alcohol consumption and khat chewing, had a higher risk of undiagnosed hypertension.51,53

The combined use of khat and alcohol creates a synergistic surge in blood pressure. Khat contains cathinone, a stimulant that triggers the sympathetic nervous system and constricts blood vessels.⁵⁴ Ethanol consumption triggers a pressor response by inducing systemic vasoconstriction. It stimulates the hypothalamic-pituitary-adrenal (HPA) axis, leading to elevated cortisol and catecholamine levels.⁵⁵ Similarly, cathinone, the primary alkaloid in khat, functions as a potent sympathomimetic agent. It augments myocardial contractility and peripheral resistance via the massive release of presynaptic norepinephrine.⁵⁶ From a pharmacodynamic perspective, alcohol may enhance the bioavailability of cathinone by increasing the permeability of the oral mucosa. This in turn, accelares its systemic absorption during mastication. This synergistic interaction precipitates a more profound hemodynamic challenge than the isolated consumption of either substance.⁵⁷ This concurrent exposure imposes significant cumulative physiological strain, increasing blood pressure.⁵⁸

Our study also showed that individuals who do not believe that consuming too much salt causes health problems were more likely to have undiagnosed hypertension than those who do have this belief. A study conducted among the Ghanaian and South African populations also indicates that about one-third of the population was unaware that a high salt diet could lead to health problems.⁵⁹ Despite this shared regional lack of awareness, Ethiopia’s policy response has been limited. Most nutrition policies in Ethiopia align with the WHO guidelines on healthy eating behaviour. However, salt reduction strategies remain notably under-addressed.60,62 While Ethiopia’s Nutrition for Growth Commitment endorsed the WHO salt reduction guidelines, the implementation remains weak.⁶³ Many individuals in rural areas lack awareness of the health risks associated with high salt intake.⁶⁴

Excessive sodium intake has a profound physiological impact. It leads to alterations in small arteries, which elevates peripheral vascular resistance. Elevated sodium levels cause vascular restoration, reducing the ability of blood vessels to dilate and finally raising blood pressure.⁶⁵

The current study showed that, as the body mass index increases, both systolic and diastolic blood pressure increase. A study conducted in Nigeria also supports this finding. The prevalence of hypertension was 4.7% among the malnourished population compared with 55.5% among normal-weight individuals.⁸ Similarly, an Ethiopian study among adults found that underweight individuals had a lower risk of hypertension than normal-weight individuals.⁹ This may be attributed to lower systemic adiposity and decreased total peripheral resistance among the underweight people compared to the overweight.⁶⁶

Regular blood pressure measurement is important to detect cases at an early stage.⁶⁷ In our study, never having had blood pressure measured before is significantly associated with undiagnosed hypertension. These findings are consistent with previous evidence from the Gambia and the United States.^{68 69} Data from the Gambian study showed that a lack of prior blood pressure screening serves as a proxy for limited health literacy and awareness; consequently in a higher risk for undiagnosed hypertension.⁶⁸ The other evidence indicated that elevated blood pressure readings during childhood and adolescence stage are associated with the development of hypertension in adulthood.⁶⁹ In line with these findings, prior Ethiopian research identified frequent screening as a preventive measure for hypertension.⁷⁰ Health-seeking behaviors, socioeconomic status, and cultural factors affected screening habits.^{5 70} In rural Ethiopia, traditional beliefs and inadequate healthcare infrastructure obstruct regular screening efforts.⁵

Methodological implications of the study

The key implication of this study is the adoption of a two-stage diagnostic strategy rather than a single-visit blood pressure measurement. Because mean blood pressure differed substantially between community level screening and the confirmation visit, first-visit measurements alone would have led to misclassification. We therefore recommend that clinicians and researchers routinely use repeated blood pressure measurements across visits to establish a diagnosis of hypertension.

Furthermore, the use of census data as a population-representative sampling frame minimizes selection bias and provides nearly unbiased estimates for the target population. This in turn enhances the external validity and generalizability of the findings. In addition to that, the application of Generalized Estimation Equation helps to provide more reliable, population-averaged estimates of the risk factors for undiagnosed hypertension.

Limitations

This study has some limitations. Firstly, some screen-positive individuals did not attend their referral at the health centre. This might have influenced the prevalence estimation. Subgroup and sensitivity analyses were made to mitigate this; however, some degree of attrition bias may remain. Secondly, despite the use of validated automated devices and extensive training of data collectors to reduce observer variability, measurement error cannot be completely ruled out. Thirdly, information on alcohol drinking, khat chewing, and cigarette smoking was obtained by interviewing the study participants. This may have resulted in social desirability bias. To mitigate this, interviews were conducted in private settings, and the confidentiality of respondents was protected.

Conclusion

According to the current study, the prevalence of newly diagnosed hypertension cases ranged from 7.7% to 14.3%. Over four-fifths (529/625 suspected cases) were newly detected through community screening, yet one-third normalized at follow-up. This showed how repeated measurements and multi-stage diagnostic protocols mitigate transient blood pressure elevations. The prevalence of previously diagnosed cases was 3.3%, underscoring needs for community-based screening in rural settings.

Older age, female sex, alcohol and/or khat use, poor knowledge of salt-related health risks, and a history of no prior blood pressure check-up were associated with undiagnosed hypertension independently.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary file.