Abstract
The authors investigated the effects of single and multiple blood pressure (BP) measurements during the same encounter on screen‐detected diabetes risk. Data for 9018 Cameroonian adults from a community‐based survey were used. Resting BP was measured three times 5 minutes apart. Logistic regressions were used to compute the odd ratio (OR) per standard deviation (SD) higher BP variables. Systolic BP, diastolic BP, and mean arterial pressure (MAP), but not pulse pressure, were related to prevalent diabetes. The highest OR (95% confidence interval [CI]) per SD higher pressure were recorded for MAP (OR, 1.16; 95% CI, 1.05–1.28) and systolic BP (OR, 1.15; 95% CI, 1.04–1.27). Estimates of the association were highest for the first, then third, and lastly the second BP measurements. Estimates from average BP measurements were not better than those from single measurement. Single BP measurement is more effective for diabetes risk screening than multiple measurements. Community‐based diabetes strategies utilizing a single measurement are simple without compromising the yield.
Higher‐than‐optimal blood pressure (BP) levels are major determinants of cardiovascular disease risk in diverse populations and settings.1, 2, 3 Diabetes mellitus and other forms of dysglycemia are important public health problems and high blood pressure (BP) has been established as a major driver of the adverse health consequences of diabetes and dysglycemia.4 Some interventions for controlling BP are also effective for controlling or reducing the risk of diabetes, and lowering BP indices are major targets of strategies for preventing or slowing the progression of diabetes complications.1, 3, 5
High BP and dysglycemia tend to coincide, and it is established in routine clinical practice that the presence of one of these abnormalities in an individual should trigger investigations for the presence of the other. As an extension of the above, in the present era where diabetes screening is increasingly advocated, status of hypertension and/or BP variables have been incorporated in multivariable models to predict the presence or future occurrence of diabetes. However, strategies for measuring BP in a routine setting and/or ascertaining the presence of hypertension varies, and can potentially be a deterrent to the adoption of multivariable models, particularly when BP measurement can be time‐consuming. For instance, there is a consensus that BP should be measured at least twice during the same encounter to establish accurate BP levels. However, whether strategies based on a single BP measurement during a clinical encounter or the average of several measurements during the same encounter has similar effects on diabetes risk prediction is still uncertain. Previous studies mostly in Caucasian and to some extent Asian populations have evaluated the impact of single measurement, average of multiple measurements of BP, or usual BP levels on the risk of diabetes mellitus, with variable findings.6 However, the value of a single BP measurement at different time points in predicting prevalent diabetes has not yet been established in developing countries and the African setting in particular.
The aim of this study was to assess and compare the effects of a single BP measurement vs multiple BP measurements at different time points during the same encounter, and their averages, on the risk of screen‐detected diabetes mellitus in urban Cameroonians.
Methods
Study Population and Settings
Participants were adult Cameroonians (age range, 15–99 years) who took part in the Cameroon Burden of Diabetes (CAMBoD) 2006–2008 community‐based survey, which has been previously described in detail.7 Briefly, this was a population‐based survey conducted in 2006 in a sample of more than 10,000 adult Cameroonians recruited from 4235 households across four ecological zones in Cameroon. The study sites were the health districts of Biyem‐Assi, Yaoundé; Cité des Palmiers, Douala; Bamenda; and Garoua. The study conforms to the Helsinki declaration and was approved by the Cameroon National Ethical Committee and the Ministry of Public Health. Informed consent was obtained from each participant before enrollment. Excluded from the study were pregnant women, individuals who did not comply with the study instructions, and psychiatric patients.
Data Collection
For all consenting participants, data were collected during household visits on age, sex, medical history, risk factors for diabetes, and behavioral factors (smoking and alcohol consumption). Physical measurements included height, weight, and waist and hip circumference assessed with standard methods. Body mass index was calculated as weight (kg)/height*height (m2).
BP Measurement
Following a 10‐minute rest with the patients in a seated position, three repeated BP measurements 5 minutes apart were performed on the right arm by the same investigators using digital home BP monitors (model UA‐767, A&D Company Limited, Tokyo, Japan) and appropriately sized cuffs during household visits. The device uses oscillometric measurement methods to acquire BP levels. Mean arterial pressure (MAP) was estimated as DBP+1/3 (SBP–DBP) and pulse pressure (PP) as SBP–DBP, with DBP as diastolic BP and SBP as systolic BP.
Diabetes Assessment
Fasting capillary glucose (FCG) was assessed using the HemoCue B‐glucose photometer (Ängelholm, Sweden) after an overnight fast. During prerecruitment sensitization visits, participants were advised on the importance of an “overnight fast” on the accuracy of their test results. Validation studies have reported very good correlation (>0.97) between glucose measurement from the HemoCue B‐glucose device and glucose measurements from several standard methods.8 Diabetes was defined using fasting capillary glucose with an abnormal FCG ≥110 mg/dL or 6.1 mmol/L.
Statistical Methods
Data were analyzed with SAS/STAT version 9.1 (SAS Institute, Cary, NC) and were restricted to participants with no history of doctor‐diagnosed diabetes and with data available on all BP measurements. Data are summarized as mean and standard deviation (SD) or number (percentages). Group comparison used chi‐square test and Student t test or nonparametric equivalents. Logistic regression models were used to assess the independent association between prevalent diabetes and each index of BP. The independent association was assessed in two ways: first by fitting a continuous predictor to obtain the odd ratio (OR) and 95% confidence interval (CI) for an SD higher level of BP variables and, second by comparing the association across fifths of BP variables. In the latter analyses, the lower fifth of BP variables was always used as reference. The log‐linearity of the associations between each BP variable and screen‐detected diabetes was explored. All models were adjusted for age and sex. In expanded multivariable models comprising age, sex, family history of diabetes, physical activity, and BMI, the area under the receiver operating characteristic curve (AUC) was used to assess and compare the ability of BP variables to discriminate between participants with and without prevalent diabetes mellitus. Receiver operating characteristic analysis (ROC) comparisons used the algorithm of DeLong and colleagues9 The likelihood ratio χ2 statistics for each pressure variable were calculated by comparing multivariate regression models with and without the BP variable of interest to measure the strength of the association with screen‐detected diabetes. We also calculated the Akaike's information criterion (AIC), the Hosmer & Lemeshow calibration chi‐square, and the accompanying P value.
Results
Baseline Characteristics of Participants
Of the 10,000 participants who took part in the survey, 982 had doctor‐diagnosed diabetes or missing data on diabetes status or one of the three BP measurements and were all excluded. Therefore, the final analytic sample comprised 9018 participants. The profile of the 3544 men and 5474 women in the final sample is described in Table 1. With the exception of fruit (P=.975) and vegetable consumption (P=.678), family history of diabetes (P=.206), and existing hypertension (P=.376), significant sex differences were apparent in the distribution of most baseline characteristics. Compared with women, men were older (35 vs 34 years, P=.0023), more likely to be ever smokers (28.0% vs 5.9%, P<.0001), to be alcohol drinkers (67.6% vs 59.6%, P<.0001), to have a high waist‐hip ratio (0.86 vs 0.82, P<.0001), to have higher BP levels regardless of the time of measurement (all P≤.002), and lower body mass index (23.7 vs 26.0 kg/m2, P<.0001). The prevalence of screen‐detected diabetes mellitus was 4.5% overall and similar in men and women (P=.930) (Table 1.
Table 1.
Baseline Characteristics of the Study Patients
| Variables | Men | Women | P Value | Total |
|---|---|---|---|---|
| No. | 3544 | 5474 | 9018 | |
| Age, y | 35 (16) | 34 (15) | .0023 | 34 (15) |
| Weight, kg | 69.0 (13.0) | 67.6 (15.8) | <.0001 | 68.2 (14.8) |
| Height, cm | 171 (8) | 161 (7) | <.0001 | 165 (9) |
| BMI, kg/m2 | 23.7 (4.1) | 26.0 (6.0) | <.0001 | 25.1 (5.5) |
| Waist circumference, cm | 81 (11) | 83 (13) | <.0001 | 82 (13) |
| Hip circumference, cm | 94 (9) | 100 (13) | <.0001 | 98 (12) |
| Waist‐hip ratio | 0.86 (0.09) | 0.82 (0.13) | <.0001 | 0.84 (0.12) |
| Known hypertension, No. (%) | 251 (7.1) | 415 (7.6) | .376 | 666 (7.4) |
| Family history of diabetes, No. (%) | 321 (9.1) | 540 (9.9) | .206 | 861 (9.6) |
| Family history of hypertension, No. (%) | 739 (20.9) | 1368 (25.0) | <.0001 | 2107 (23.4) |
| BP variables, mm Hg | ||||
| First BP measurements | ||||
| Systolic BP | 129 (20) | 124 (22) | <.0001 | 126 (21) |
| Diastolic BP | 76 (14) | 76 (14) | .002 | 76 (14) |
| Mean arterial pressure | 94 (14) | 92 (15) | <.0001 | 93 (15) |
| Pulse pressure | 53 (14) | 48 (15) | <.0001 | 50 (15) |
| Second BP measurements | ||||
| Systolic BP | 126 (19) | 120 (21) | <.0001 | 122 (20) |
| Diastolic BP | 75 (13) | 74 (14) | .002 | 75 (14) |
| Mean arterial pressure | 92 (14) | 90 (15) | <.0001 | 91 (15) |
| Pulse pressure | 50 (13) | 46 (14) | <.0001 | 48 (14) |
| Third BP measurements | ||||
| Systolic BP | 125 (19) | 119 (20) | <.0001 | 121 (20) |
| Diastolic BP | 75 (13) | 73 (13) | <.0001 | 74 (13) |
| Mean arterial pressure | 91 (14) | 88 (15) | <.0001 | 90 (14) |
| Pulse pressure | 50 (13) | 45 (13) | <.0001 | 47 (13) |
| Fasting blood glucose, mg/dL | 90 (26) | 92 (18) | .0031 | 91 (22) |
| Fruit consumption, d/wk | 2 (1–4) | 2 (1–4) | .975 | 2 (1–4) |
| Vegetable consumption, d/wk | 3 (2–5) | 3 (2–4) | .678 | 3 (2–4) |
| Ever smokers, No. (%) | 989 (28.0) | 321 (5.9) | <.0001 | 1310 (14.5) |
| Ever consumes alcohol, No. (%) | 2384 (67.6) | 3259 (59.6) | <.0001 | 5643 (62.7) |
| Physical activity at leisure, No. (%) | 2914 (83.1) | 5082 (93.8) | <.0001 | 7996 (89.6) |
| Education | ||||
| None | 355 (10.0) | 1044 (19.1) | <.0001 | 1399 (15.5) |
| Primary | 1497 (42.2) | 2532 (46.3) | 4029 (44.7) | |
| Secondary | 1329 (37.5) | 1620 (29.6) | 2949 (32.7) | |
| Tertiary | 363 (10.2) | 278 (5.1) | 641 (7.1) | |
| Screen‐detected diabetes, No. (%) | 160 (4.5) | 245 (4.5) | .930 | 405 (4.5) |
Abbreviation: BMI, body mass index; BP, blood pressure. Values are expressed as numbers (percentages) and means (standard deviations) or median (minimum–maximum).
Association of Different BP Measurements With Prevalent Diabetes
In logistic regression models adjusted for sex and age, there was a continuous association between single BP measurements (first, second, or third), their average (mean first and second; mean second and third; and mean first, second, and third) and screen‐detected diabetes. Across different BP variables measured at the same time point, the highest point estimates of the association (odds ratio) per SD higher pressure were always recorded with MAP or SBP and the lowest always with PP. Across measurements of the same pressure variable at different time points, point estimates of the associations were mostly higher for the first measurement, then the third and lastly the second measurement. Furthermore, estimates of the associations from various averages of pressure variables from different measurements were mostly similar for each pressure variable, and not appreciably different from estimates obtained using the first measurement alone. The expected dose‐response association was observed across fifths of all BP variables at different time points, as well as their averages. These associations were always log‐linear (all P<.02 for log‐linear trend) (Table 2).
Table 2.
Risk of Screen‐Detected Diabetes Across Fifths of BP Variables
| Measurement | Systolic BP | Diastolic BP | Mean Arterial Pressure | Pulse Pressure | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SBP | Fifths | Median | OR (95%CI) | Fifths | Median | OR (95%CI) | Fifths | Median | OR (95%CI) | Fifths | Median | OR (95%CI) |
| 1 | 1 | 104 | 1.00 (0.77–1.29) | 1 | 60 | 1.00 (0.77–1.29) | 1 | 77 | 1.00 (0.77–1.29) | 1 | 34 | 1.00 (0.78–1.29) |
| 2 | 114 | 1.14 (0.90–1.43) | 2 | 68 | 1.00 (0.78–1.27) | 2 | 84 | 1.12 (0.88–1.42) | 2 | 42 | 1.16 (0.92–1.47) | |
| 3 | 123 | 1.14 (0.90–1.43) | 3 | 74 | 1.12 (0.90–1.38) | 3 | 90 | 1.27 (1.02–1.57) | 3 | 58 | 1.24 (0.99–1.55) | |
| 4 | 132 | 1.21 (0.96–1.52) | 4 | 81 | 1.20 (0.96–1.49) | 4 | 97 | 1.19 (0.95–1.49) | 4 | 54 | 1.36 (1.10–1.69) | |
| 5 | 152 | 1.52 (1.22–1.89) | 5 | 92 | 1.28 (1.03–1.59) | 5 | 111 | 1.40 (1.12–1.75) | 5 | 68 | 1.35 (1.09–1.68) | |
| SD | 21.0 | 1.15 (1.04–1.27) | SD | 13.7 | 1.13 (1.03–1.25) | SD | 15.0 | 1.16 (1.05–1.28) | SD | 14.9 | 1.08 (0.98–1.19) | |
| P trend | .0002 | P trend | .001 | P trend | .0005 | P trend | .006 | |||||
| 2 | 1 | 101 | 1.00 (0.77–1.29) | 1 | 59 | 1.00 (0.77–1.30) | 1 | 75 | 1.00 (0.78–1.27) | 1 | 33 | 1.00 (0.79–1.27) |
| 2 | 112 | 1.24 (0.98–1.57) | 2 | 67 | 1.01 (0.80–1.27) | 2 | 83 | 0.86 (0.64–1.07) | 2 | 41 | 1.02 (0.81–1.28) | |
| 3 | 119 | 1.28 (1.02–1.61) | 3 | 73 | 1.06 (0.85–1.32) | 3 | 89 | 1.06 (0.85–1.33) | 3 | 46 | 1.08 (0.85–1.38) | |
| 4 | 128 | 1.30 (1.05–1.62) | 4 | 80 | 1.09 (0.87–1.37) | 4 | 95 | 1.11 (0.89–1.39) | 4 | 52 | 1.22 (0.98–1.52) | |
| 5 | 147 | 1.46 (1.16–1.84) | 5 | 92 | 1.35 (1.08–1.68) | 5 | 108 | 1.26 (1.01–1.57) | 5 | 64 | 1.25 (1.01–1.55) | |
| SD | 20.2 | 1.13 (1.03–1.25) | SD | 13.6 | 1.12 (1.01–1.23) | SD | 14.7 | 1.14 (0.81–1.58) | SD | 13.8 | 1.07 (0.98–1.18) | |
| P trend | .0006 | P trend | .0007 | P trend | .0004 | P trend | .019 | |||||
| 3 | 1 | 101 | 1.00 (0.78–1.28) | 1 | 59 | 1.00 (0.78–1.28) | 1 | 74 | 1.00 (0.78–1.28) | 1 | 33 | 1.00 (0.80–1.25) |
| 2 | 111 | 0.86 (0.66–1.12) | 2 | 67 | 0.97 (0.76–1.24) | 2 | 81 | 0.97 (0.76–1.24) | 2 | 40 | 0.82 (0.63–1.07) | |
| 3 | 118 | 1.27 (1.01–1.59) | 3 | 72 | 0.95 (0.75–1.21) | 3 | 88 | 1.02 (0.82–1.29) | 3 | 45 | 0.96 (0.77–1.21) | |
| 4 | 126 | 1.20 (0.97–1.49) | 4 | 79 | 1.07 (0.86–1.33) | 4 | 94 | 1.03 (0.82–1.29) | 4 | 52 | 1.20 (0.96–1.50) | |
| 5 | 144 | 1.37 (1.09–1.71) | 5 | 90 | 1.22 (0.98–1.51) | 5 | 107 | 1.31 (1.05–1.63) | 5 | 63 | 1.19 (0.96–1.47) | |
| SD | 19.9 | 1.15 (1.04–1.27) | SD | 13.4 | 1.10 (1.00–1.22) | SD | 14.5 | 1.14 (1.03–1.26) | SD | 13.5 | 1.11 (1.01–1.22) | |
| P trend | .0001 | P trend | .003 | P trend | .001 | P trend | .012 | |||||
| Mean of 1, 2, and 3 | 1 | 103 | 1.00 (0.78–1.28) | 1 | 60 | 1.00 (0.78–1.28) | 1 | 76 | 1.00 (0.78–1.29) | 1 | 35 | 1.00 (0.78–1.28) |
| 2 | 113 | 0.94 (0.73–1.21) | 2 | 68 | 0.84 (0.66–1.08) | 2 | 83 | 0.97 (0.76–1.24) | 2 | 41 | 1.31 (1.05–1.64) | |
| 3 | 120 | 1.15 (0.91–1.44) | 3 | 73 | 0.97 (0.77–1.22) | 3 | 89 | 1.18 (0.94–1.47) | 3 | 46 | 1.10 (0.87–1.40) | |
| 4 | 128 | 1.28 (1.04–1.59) | 4 | 80 | 1.14 (0.92–1.42) | 4 | 95 | 1.21 (0.98–1.50) | 4 | 52 | 1.45 (1.17–1.80) | |
| 5 | 147 | 1.34 (1.07–1.69) | 5 | 91 | 1.22 (0.99–1.52) | 5 | 109 | 1.34 (1.07–1.68) | 5 | 64 | 1.39 (1.12–1.74) | |
| SD | 19.6 | 1.15 (1.04–1.27) | SD | 12.7 | 1.13 (1.02–1.25) | SD | 14.1 | 1.15 (1.04–1.28) | SD | 12.6 | 1.10 (1.00–1.21) | |
| P trend | .0002 | P trend | .0004 | P trend | .0002 | P trend | .006 | |||||
| Mean of 1 and 2 | 1 | 103 | 1.00 (0.77–1.29) | 1 | 60 | 1.00 (0.78–1.29) | 1 | 76 | 1.00 (0.77–1.29) | 1 | 35 | 1.00 (0.78–1.28) |
| 2 | 113 | 0.99 (0.77–1.27) | 2 | 68 | 0.99 (0.78–1.25) | 2 | 83 | 1.13 (0.89–1.44) | 2 | 41 | 1.37 (1.11–1.69) | |
| 3 | 120 | 1.28 (1.03–1.58) | 3 | 74 | 1.06 (0.84–1.33) | 3 | 89 | 1.24 (1.00–1.55) | 3 | 46 | 0.88 (0.67–1.15) | |
| 4 | 129 | 1.27 (1.02–1.58) | 4 | 80 | 1.09 (0.87–1.36) | 4 | 96 | 1.19 (0.95–1.48) | 4 | 52 | 1.40 (1.13–1.72) | |
| 5 | 148 | 1.38 (1.10–1.74) | 5 | 91 | 1.32 (1.06–1.65) | 5 | 109 | 1.43 (1.14–1.79) | 5 | 64 | 1.33 (1.07–1.65) | |
| SD | 20.0 | 1.15 (1.04–1.27) | SD | 12.9 | 1.14 (1.03–1.26) | SD | 14.4 | 1.16 (1.04–1.28) | SD | 13.1 | 1.08 (0.98‐1.19) | |
| P trend | .0002 | P trend | .0008 | P trend | .0004 | P trend | .022 | |||||
| Mean of 2 and 3 | 1 | 102 | 1.00 (0.78–1.28) | 1 | 59 | 1.00 (0.77–1.29) | 1 | 75 | 1.00 (0.78–1.28) | 1 | 34 | 1.00 (0.79–1.27) |
| 2 | 111 | 0.94 (0.73–1.21) | 2 | 67 | 0.91 (0.72–1.16) | 2 | 82 | 0.91 (0.71–1.16) | 2 | 40 | 1.15 (0.91–1.46) | |
| 3 | 118 | 1.18 (0.94–1.47) | 3 | 73 | 1.04 (0.83–1.31) | 3 | 88 | 1.10 (0.88–1.38) | 3 | 45 | 1.10 (0.88–1.38) | |
| 4 | 126 | 1.21 (0.97–1.50) | 4 | 79 | 1.17 (0.95–1.45) | 4 | 95 | 1.17 (0.94–1.45) | 4 | 51 | 1.37 (1.11–1.70) | |
| 5 | 144 | 1.36 (1.09–1.71) | 5 | 90 | 1.26 (1.01–1.57) | 5 | 108 | 1.30 (1.04–1.63) | 5 | 63 | 1.30 (1.05–1.63) | |
| SD | 19.6 | 1.15 (1.04–1.27) | SD | 12.9 | 1.12 (1.01–1.24) | SD | 14.3 | 1.14 (1.03–1.26) | SD | 12.7 | 1.10 (1.00–1.21) | |
| P trend | .0003 | P trend | .0005 | P trend | .0003 | P trend | .01 | |||||
All analyses were adjusted for age and sex. For each blood pressure (BP) variable, the odds ratio (OR) and 95% confidence interval (CI) for a standard deviation (SD) higher level of the variable is shown, together with the P value for the log‐linearity of the association (P trend).
Prediction of Prevalent Diabetes by Multivariable Models Comprising Different BP Measurements
When models with BP variables and covariates were compared with models with covariates only, the difference in the likelihood ratio χ2 test was in favor of a stronger association between single measurement (first) of pressure variable and prevalent diabetes, although the difference with mean measurements of BP was marginal (Table 3).
Table 3.
Prediction of Prevalent Diabetes Mellitus
| Variable | Measurement | AUC (CI) | AIC | Delta Likelihood Ratio χ2 | Calibration χ2 (P Value) |
|---|---|---|---|---|---|
| SBP | 1 | 0.600 (0.570–0.630) | 3198.095 | 4.21 | 14.69 (.065) |
| 2 | 0.602 (0.573–0.632) | 3199.211 | 3.09 | 12.28 (.129) | |
| 3 | 0.602 (0.572–0.632) | 3198.415 | 3.89 | 12.27 (.129) | |
| Mean of 1, 2, and 3 | 0.602 (0.572–0.632) | 3198.202 | 4.10 | 12.55 (.128) | |
| Mean of 1 and 2 | 0.602 (0.572–0.631) | 3198.368 | 3.93 | 11.83 (.159) | |
| Mean of 2 and 3 | 0.602 (0.573–0.632) | 3198.618 | 3.68 | 15.11 (.057) | |
| Global P | .791 | ||||
| DBP | 1 | 0.600 (0.570–0.630) | 3200.349 | 1.95 | 15.65 (.048) |
| 2 | 0.600 (0.570–0.630) | 3201.281 | 1.02 | 19.89 (.011) | |
| 3 | 0.600 (0.570–0.630) | 3201.909 | 0.39 | 18.61 (.017) | |
| Mean of 1, 2, and 3 | 0.600 (0.570–0.630) | 3201.080 | 1.22 | 15.45 (.051) | |
| Mean of 1 and 2 | 0.600 (0.570–0.630) | 3200.635 | 1.67 | 20.00 (.010) | |
| Mean of 2 and 3 | 0.600 (0.570–0.630) | 3201.555 | 0.75 | 19.19 (.014) | |
| Global P | .930 | ||||
| MAP | 1 | 0.600 (0.570–0.630) | 3198.894 | 3.41 | 11.76 (.162) |
| 2 | 0.601 (0.571–0.631) | 3200.204 | 2.10 | 17.31 (.027) | |
| 3 | 0.601 (0.571–0.630) | 3200.584 | 1.72 | 16.67 (.034) | |
| Mean of 1, 2, and 3 | 0.601 (0.571–0.631) | 3199.669 | 2.63 | 14.13 (.078 | |
| Mean of 1 and 2 | 0.601 (0.571–0.631) | 3199.339 | 2.96 | 12.80 (.119) | |
| Mean of 2 and 3 | 0.601 (0.571–0.631) | 3200.267 | 2.03 | 14.40 (.072) | |
| Global P | .995 | ||||
| PP | 1 | 0.600 (0.571–0.630) | 3200.032 | 2.27 | 19.46 (.013) |
| 2 | 0.603 (0.574–0.633) | 3200.039 | 2.26 | 20.28 (.009) | |
| 3 | 0.604 (0.574–0.633) | 3197.736 | 4.57 | 10.62 (.224) | |
| Mean of 1, 2, and 3 | 0.603 (0.574–0.633) | 3198.654 | 3.65 | 20.86 (.007) | |
| Mean of 1 and 2 | 0.603 (0.573–0.632) | 3199.587 | 2.71 | 23.29 (.003) | |
| Mean of 2 and 3 | 0.604 (0.575–0.634) | 3198.512 | 3.79 | 14.74 (.064) | |
| Global P | .340 |
Abbreviations: AIC, Akaike information criterion; AUC, area under the receiver operating characteristic curve; CI, confidence interval; DBP, diastolic blood pressure; MAP, mean arterial pressure; PP, pulse pressure; SBP, systolic blood pressure. Models are adjusted for age, sex, family history of diabetes, physical activity, and body mass index.
Based on AIC comparison, models with the first measurement of SBP, DBP, and MAP were the best performing, while the model with the average of the second and third measurements was the best for PP (Table 3). However, the discriminating capability as appreciated by the AUC was not appreciably different between models with single and averages of multiple BP measurements (all P≥.34) for all AUC comparisons (Table 3).
Discussion
We compared BP measurements at different time points and their averages, for screening individuals with prevalent undiagnosed diabetes in the Cameroonian adult population. The current analyses indicated that single and multiple BP measurements, including SBP, DBP, MAP, and to a lesser extent PP were continuously associated with increased risk of prevalent screen‐detected diabetes. The highest point estimates of the association across measurements of the same BP variable at different time points were noted for the first, then third, and finally the second measurement. Furthermore, estimates of associations from the different measurements of the means of the BP variables were similar for each pressure variable and not appreciably different from estimates obtained using the first BP measurement alone. These associations appeared to be log linear, with an SD higher level of each BP measurement, contributing the same range of effect sizes as indicated by the overlapping confidence interval about the effect estimates. Multivariable models containing classical diabetes risk factors and each BP variable (either from a single measurement of the average of several measurements) had similar discriminatory capability for prevalent diabetes.
Accurate acquisition of BP levels as well as many other biological parameters is subject to adequate control of measurement errors, which is the variation between the observed value and the so‐called usual value, the equivalent of the individual's long‐term value. Measurement error is further segregated into “technical error” due, for instance, to an imprecise measurement device (eg, BP machine), and individual's biological variation over time. One implication of measurement errors is that it can introduce biases into the estimation of regression coefficients, often in the direction of underestimating the true effect estimate; hence, the appellation regression dilution bias.10 This underestimation of the effect tends to be small when the average of two or more measurements of the variables of interest are used, and at least in part motivates the recommendation of relying on multiple measurements of BP in surveys and studies on hypertension and nonoptimal BP.11
Approximating as much as possible the true levels of a single biological variable has relevance when the obtained value will be used for decision‐making regarding ordering new tests, starting individuals on treatments, or withholding such treatments, where the balance of risks, benefits, cost, and acceptability have to be accounted for. It has been suggested that when the intended purpose is to test the hypothesis of a linear relationship between two variables, and not to estimate the effect size of this association, more accurate measurement of the variables is not particularly relevant. Accurate measurement is also likely less relevant even when the intended purpose is to estimate the effect size of the relationship in multivariable regression models, particularly when the contribution of the targeted variable to the overall risk is small, and when such a variable is tested alongside more powerful predictors of the outcomes of interest. Our analyses suggest that this is the case for BP in multivariable models for diabetes risk prediction. In minimally adjusted models in our study, the effect size of the association of pressure variables with diabetes risk was modest, with each SD higher pressure level, conferring at most a 15% higher risk of screen‐detected diabetes, with no sizable variation between single and multiple measurements. In the presence of five other common risk factors for diabetes, the association of pressure variables with diabetes risk was very weak, with a single BP and the average of three measurements having the same strength. As a consequence, the overall performance of resulting models in terms of discrimination was similar. Altogether, our analysis suggests that improving the accuracy of BP assessment through multiple measurements has no impact on prevalent diabetes risk screening.
We are not aware of an existing study that has compared the effects of the same BP variables measured at different time points on the risk of diabetes. However, few studies have compared the predictive utility of different BP variables measured at the same time point in relation to diabetes risk12, 13, 14, 15 or the risk of related complications.1, 16, 17 In line with our findings, these studies have been consistent in suggesting SBP to be one of the most effective discriminators of the outcomes risk among pressure variables. But the magnitude of the effects appears to be more important in relation to diabetes complications than to diabetes presence or occurrence. Our findings have implications for the implementation of diabetes risk screening strategies, particularly in community‐based settings. In such settings, the overall aim would be to use simple approaches to detect people who are more likely to be diagnosed with diabetes (for instance, if tested via blood tests in the healthcare setting). For this approach, simplified BP measurement (single measurement as opposed to multiple measurements) will facilitate the application of validated diabetes risk models based on noninvasively measured predictors. With the increased availability of automated BP measurement devices, it is conceivable that simplified BP measurement as well as measurement of other common risk factors for diabetes will facilitate self‐screening and self‐referral of individuals using noninvasive diabetes risk models. We found in a previous review that the uptake of those models was still at the early infancy in Africa, and that issues with measurement of predictors included in those models could affect their applicability in this setting.18
Study Limitations and Strengths
The present study has one major limitation. Screen‐detected diabetes was based on fasting glucose only, and in the absence of oral glucose tolerance test, it may be possible that participants with diabetes based on 2‐hour glucose would have been misclassified as nondiabetic. It is unlikely that they would have been in sufficient numbers to affect the conclusions of the study. Indeed, there is no indication that BP variables are associated in differential ways to fasting glucose–diagnosed and 2‐hour glucose–based diabetes. As such, a possible effect of our failure to account for those with diabetes based on 2‐hour glucose is the dilution of the true effect of the association of pressure variables with diabetes risk. The rather young age of our participants, reflecting the young age of the background population, would also translate into a lower prevalence of diabetes in the youngest segment of the population. Because this segment is also more likely to have lower BP levels, it is possible that our ability to carefully examine the diabetes‐BP relationship was limited in the lower tail of BP distribution. The present study also has major strengths, ranging from its multicenter settings with wide coverage, its relatively large representative sample, direct measurements of BP, and the use of advanced analytical techniques to investigate the associations and compare the predictive values.
Conclusions
The common BP variables are weakly associated with prevalent diabetes risk, with no indication that better approximation of the “usual BP level” through averaging multiple measurements does better than a single measurement in discriminating the risk of diabetes. Furthermore, SBP appears to be the most effective single BP variable for diabetes risk, with the first measurement performing equally as well as the average of three consecutive measurements during the same encounter. Our findings support the use of simplified BP measurement in community‐based approaches to diabetes risk screening using multivariable models based on noninvasively measured predictors including BP variables.
Disclosure
The authors declare no conflict of interest.
Acknowledgments
We acknowledge the Cameroon Burden of Diabetes (CAMBoD) survey team members for providing us with logistic support. We are also thankful to all participants in the CAMBoD survey for their cooperation. The CAMBoD projects were sponsored by World Diabetes Foundation grants WDF 02‐016 and WDF 05‐117.
J Clin Hypertens (Greenwich). 2016;18:864–870. DOI: 10.1111/jch.12774. © 2016 Wiley Periodicals, Inc.
References
- 1. Kengne AP, Czernichow S, Huxley R, et al. Blood pressure variables and cardiovascular risk: new findings from ADVANCE. Hypertension. 2009;54:399–404. [DOI] [PubMed] [Google Scholar]
- 2. Lawes CM, Bennett DA, Parag V, et al. Blood pressure indices and cardiovascular disease in the Asia Pacific region: a pooled analysis. Hypertension. 2003;42:69–75. [DOI] [PubMed] [Google Scholar]
- 3. Lawes CM, Rodgers A, Bennett DA, et al. Blood pressure and cardiovascular disease in the Asia Pacific region. J Hypertens. 2003;21:707–716. [DOI] [PubMed] [Google Scholar]
- 4. Kengne AP, Libend CN, Dzudie A, et al. An assessment of discriminatory power of office blood pressure measurements in predicting optimal ambulatory blood pressure control in people with type 2 diabetes. Pan Afr Med J. 2014;19:231. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Rationale and design of the ADVANCE study: a randomised trial of blood pressure lowering and intensive glucose control in high‐risk individuals with type 2 diabetes mellitus. Action in Diabetes and Vascular Disease: PreterAx and DiamicroN Modified‐Release Controlled Evaluation. J Hypertens Suppl. 2001;19:S21–S28. [PubMed] [Google Scholar]
- 6. Emdin CA, Anderson SG, Woodward M, Rahimi K. Usual blood pressure and risk of new‐onset diabetes: evidence from 4.1 million adults and a meta‐analysis of prospective studies. J Am Coll Cardiol. 2015;66:1552–1562. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Mbanya VN, Kengne AP, Mbanya JC, Akhtar H. Body mass index, waist circumference, hip circumference, waist‐hip‐ratio and waist‐height‐ratio: which is the better discriminator of prevalent screen‐detected diabetes in a Cameroonian population? Diabetes Res Clin Pract. 2015;108:23–30. [DOI] [PubMed] [Google Scholar]
- 8. Stork AD, Kemperman H, Erkelens DW, Veneman TF. Comparison of the accuracy of the HemoCue glucose analyzer with the Yellow Springs Instrument glucose oxidase analyzer, particularly in hypoglycemia. Eur J Endocrinol. 2005;153:275–281. [DOI] [PubMed] [Google Scholar]
- 9. DeLong ER, DeLong DM, Clarke‐Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44:837–845. [PubMed] [Google Scholar]
- 10. Berglund L. Regression dilution bias: tools for correction methods and sample size calculation. Ups J Med Sci. 2012;117:279–283. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. MacMahon S, Peto R, Cutler J, et al. Blood pressure, stroke, and coronary heart disease. Part 1, Prolonged differences in blood pressure: prospective observational studies corrected for the regression dilution bias. Lancet. 1990;335:765–774. [DOI] [PubMed] [Google Scholar]
- 12. Conen D, Ridker PM, Mora S, et al. Blood pressure and risk of developing type 2 diabetes mellitus: the Women's Health Study. Eur Heart J. 2007;28:2937–2943. [DOI] [PubMed] [Google Scholar]
- 13. Hsu CH, Chang JB, Liu IC, et al. Mean arterial pressure is better at predicting future metabolic syndrome in the normotensive elderly: a prospective cohort study in Taiwan. Prev Med. 2015;72:76–82. [DOI] [PubMed] [Google Scholar]
- 14. Janghorbani M, Amini M. Comparison of systolic and diastolic blood pressure with pulse pressure and mean arterial pressure for prediction of type 2 diabetes: the Isfahan Diabetes Prevention Study. Endokrynol Pol. 2011;62:324–330. [PubMed] [Google Scholar]
- 15. Smulyan H, Safar ME. The diastolic blood pressure in systolic hypertension. Ann Intern Med. 2000;132:233–237. [DOI] [PubMed] [Google Scholar]
- 16. Choukem SP, Dzudie A, Dehayem M, et al. Comparison of different blood pressure indices for the prediction of prevalent diabetic nephropathy in a sub‐Saharan African population with type 2 diabetes. Pan Afr Med J. 2012;11:67. [PMC free article] [PubMed] [Google Scholar]
- 17. Dzudie A, Choukem SP, Dehayem MY, Kengne AP. Blood pressure variables and prevalent electrocardiographic left ventricular hypertrophy in sub‐Saharan African individuals with type 2 diabetes. J Diabetes. 2012;4:424–431. [DOI] [PubMed] [Google Scholar]
- 18. Mbanya V, Hussain A, Kengne AP. Application and applicability of non‐invasive risk models for predicting undiagnosed prevalent diabetes in Africa: a systematic literature search. Prim Care Diabetes 2015;9:317–329. [DOI] [PubMed] [Google Scholar]