OUP user menu

Selective recording in blood pressure readings may increase subsequent mortality

DOI: http://dx.doi.org/10.1093/qjmed/95.9.571 571-577 First published online: 1 September 2002


Background: Rounding blood pressure (BP) to the nearest 10 mmHg (terminal digit preference) and selecting for particular values near treatment cut‐offs (number preference) have both been previously described. Both reduce measurement accuracy, and may have consequences for treatment and survival.

Aim: To check for number preference in screening for hypertension, and whether this influenced subsequent mortality.

Design: Prospective case‐control screening study.

Methods: In the General Practice Hypertensive Study Group (GPHSG), prospective case control study patients (n=23 574) were screened on one occasion for high phase‐IV diastolic BP (DBP4) (≥90 mmHg). Identified cases were matched with normotensive controls for age, sex, date of screen and ethnic group, and were registered for mortality follow‐up (n=6310). Patients with a high DBP4 had two further readings, and were treated if it remained elevated.

Results: For DBP4 terminal digit, ‘0’ was over‐represented (28.2% vs. 20%), and the number ‘88’ was over‐represented in both men and women. There was an excess adjusted death rate for females with DBP4 88–89 mmHg vs. 90–99 mmHg for both cardiovascular (RR 2.56, 95%CI 1.43–4.56, p=0.0015) and all‐cause (1.56, 95%CI 1.06–2.29, p=0.023) mortality. For males, the corresponding rates were non‐significantly reduced: cardiovascular RR 0.69, 95%CI 0.42–1.14, p=0.15; all‐cause RR 0.93, 95%CI 0.68–1.27, p=0.64.

Discussion: The quality of BP measurements should be monitored both in research studies and in clinical practice as part of clinical governance procedures.


It is known that doctors and nurses in primary and secondary care are liable to make errors in measuring blood pressure (BP) when using mercury sphygmomanometers and aneroid manometers.1–,3 These include terminal digit preference (where terminal digit zero is usually recorded more than expected)1,,4 and single‐number preference (where a particular value at or near the treatment cut‐off point is recorded more frequently than expected).5,,6 The consequences of these errors have not been described. We hypothesize that such bias may lead to errors in the classification of a patient into or out of a treatment group, with potential consequences for their risk of cardiovascular death. To test this hypothesis, we analysed data from a prospective study based in general practice, the General Practice Hypertension Study Group (GPHSG). We report the terminal digit preference and single number preference in the screening BP readings, and their relationships with patient characteristics and mortality over 23 years.


The GPHSG was formed in 1974, when eight General Practices decided to screen for hypertension in their patients and to examine their survival in a longitudinal study. Each patient between the ages of 18 and 65 years had a single BP measurement. Patients with a diastolic phase IV reading (DBP4) of <90 mmHg were classified as ‘normotensives' and not recalled. Those patients whose DBP4 was ≥90 mmHg had two further readings, except patients who were already recognized to have hypertension by their practice (‘known hypertensives’). If the second or third reading was ≥90 mmHg, the patient was classified as a ‘newly diagnosed hypertensive’, and treatment was commenced. If neither of them was ≥90 mmHg, the patient was classified as ‘transient hypertensive’. Some 95% of the target population was screened, and the methods have been reported elsewhere.7 Systolic BP (SBP) and diastolic phase V BP (DBP5) were also recorded, but did not affect the classification. Regular investigators' meetings were held, and random zero sphygmomanometers were issued to all practices, with the instruction that they were to be used for all BP readings. However, there was no formal monitoring of the quality of BP data.

Seven practices participated in a follow‐up survival study, and the data presented in this paper reflect these practices only. All patients with known hypertension, new hypertension or transient hypertension were matched with normotensive controls for age, sex, date of screen (within one year) and ethnic group. Both cases and controls were entered into the longitudinal study and registered for mortality follow‐up with the NHS central registry. Mortality rates to 31 May 1989 have been published previously.8 The present study includes deaths to 1 September 1997.


For the distributions of BP values, and for the analysis of terminal digits by GP practice, we examined data from all patients screened, and calculated mean, median and mode conventionally. For the analysis of mean age at the time of screening, we excluded patients known to be hypertensive before screening.

All death‐rate analyses were based on the cases (n=3191) and their matched controls (n=3119) only and patients known to be hypertensive before screening were excluded. Analysis of survival time was based on the Cox proportional hazards method.9 This is a technique for applying multivariate analysis to examine the effect of covariates on survival time. Graphical illustrations of survival time are calculated using the same technique. SPSS for Windows was used. We compared patients with a particular ‘number preference’ with patients with higher and lower pressures, to test the effect of screening bias on mortality. We used systolic BP to adjust for the ‘true’ level of blood pressure, since it was not affected by number preference (although it did show terminal digit preference). Age, BMI, smoking status were included in the model as covariates, since these were each significantly associated with survival (BMI was significantly associated for females only).


Terminal digits of screening readings

There was an excess of terminal digit zero in each of the screening readings for phase IV diastolic (0, 28.2%; 2, 15.0%; 4, 15.7%; 6, 17.4%; 8, 20.1%; 5, 1.8%; other odd numbers, 1.7%), phase V diastolic (0, 31.4%; 8, 16.0%) and for systolic BP (0, 30.4%; 8, 17.6%). (Table 1) Participating practices varied widely in their zero preference, with two practices showing low zero preference (18.0%, 20.2%) four with high zero preference (25.9%, 28.3%, 29.3%, 33.4%) and one with very high zero preference (49.2%). Of the 23 578 patients screened, 2824 (12%) had terminal zeros in every one of their diastolic phase IV, phase V and systolic screening readings. This compares with an expected rate of <1% or, allowing for the terminal zero preference in this study, 3%.

View this table:
Table 1

Terminal digit preference for the first diastolic (phase IV, DBP4) reading in subjects not known to have hypertension

Practice n Terminal digit (%)DBP4=88 or 89 mmHg (%)
  • The expected percentage for each even terminal digit is ∼20%, or slightly less, as occasional odd numbers are recorded. The prevalence of DBP4=88 or 89 mmHg should have been 5.5%, based on a normal distribution with the same mean and SD.

Number preference

Figure 1a shows the distribution of DBP4 for the first screening reading. There was an excess of readings at 88 mmHg over that expected, some excess of 86 mmHg and a marked deficit of 90, 92 and 94 mmHg. The percentage 88–89 mmHg in men was 6.6%, and that in women, 6.0%. Figure 1b and c show that there was also preference for 88 mmHg1 for the second and third readings, respectively. There was no ‘88’ or ‘90’ preference for DBP5 readings in the six surgeries that recorded them (Figure 2a), and there appeared to be no particular preference for any systolic BP reading (Figure 2b). The distribution of DBP4 in patients known to be hypertensive before the study is shown in Figure 2c. There was no preference for 88 mmHg.

In 27% of the screening readings (6262 readings), DBP4 and DBP5 were identical. In one Practice, 1112/1473 (75%) of the BP5 readings were recorded as unobtainable.

Figure 3 shows the average age of each patient at each value of DBP4. Males at 90 mmHg were older at 47.6 years than those with DBP4 of 88 mmHg (41.4 years). The difference for women was smaller (45.3 vs. 43.3 years).

Figure 1.

Distribution of diastolic phase IV diastolic blood pressure (DBP4) at a the first screening visit, b the second screening visit, c the third screening visit. All readings show an excess of ‘88’ readings and on the first visit, a deficit of 90, 92 and 94 mmHg.

Figure 2.

Distribution of blood pressure in the first screening reading for a diastolic phase V blood pressure (DBP5) in six of the seven surgeries, b systolic blood pressure, c diastolic phase IV blood pressure for patients known to have hypertension prior to the study. Number preference is not marked.

Figure 3.

Mean age at entry to the study at each level of first screening reading of diastolic phase IV blood pressure (DBP4) in a female and b male patients not previously known to have hypertension (with 95%CIs and linear regression line). In males, there was a marked increase in age with a DBP4 of 90 mmHg.

Mortality rates

Table 2 shows the number of deaths to 1 September 1997 from cardiovascular and all causes by the value of DBP4. It shows an excess death rate for females at initial DBP4 of 88–89 mmHg but not for males.

Cardiovascular death

In an analysis including both men and women, and after adjusting for systolic BP, age, BMI and smoking status, there was a significant interaction between sex and DBP4 band (comparison of bands 88–89 vs. 90–99 mmHg, p=0.0009). A separate analysis for females alone showed a significantly reduced survival for the 88–89 mmHg band in comparison with both the 70–87 mmHg and 90–99 mmHg bands (Table 3). The relative risk for 88–89 mmHg after adjustment was 1.87 (95%CI 1.16–3.03, p=0.01) in comparison with the 70–87 mmHg band and 2.56 (95%CI 1.43–4.56, p=0.0015) in comparison with the 90–99 mmHg band. The corresponding relative risks in men (0.79, 95%CI 0.49–1.29, p=0.34, and 0.69, 95%CI 0.42–1.14, p=0.15) did not achieve statistical significance and were in the opposite direction to the findings in women (Table 3). Figure 4 shows the survival graphs for males and females by DBP4 band.

View this table:
Table 2

Cardiovascular and total mortality rates of five bands of diastolic blood pressure and adjusted in a Cox regression for age, body mass index, systolic BP and smoking status (excluding patients previously known to have hypertension)

DBP4 (mmHg)…0–6970–8788–8990–99>100
n 2261127166605635
Total patient years in study42792131229411165211672
CVD deaths1364242570
Rate per 1000 patients years3.
Total deaths371984475140
Rate per 1000 patient years8.659.2914.966.4411.99
n 2281166197698692
Total patient years in study41312094934271257512135
CVD deaths311281988126
Rate per 1000 patient years7.506.115.547.0010.38
Total deaths5830753179220
Rate per 1000 patient years14.0414.6515.4714.2318.13

View this table:
Table 3

Relative risk (RR) of cardiovascular mortality or total mortality (with and without adjustment for age, body mass index, systolic pressure and smoking status) for men and women for DBP4 88–89 vs. 70–87 mmHg and vs. 90–99 mmHg

RR 88–89 vs. 70–87 mmHgRR 88–89 vs. 90–99 mmHg
CVD deaths2.781.873.892.56
p 0.00000.010.00000.0015
Total deaths1.631.232.371.56
p 0.0030.220.00000.023
CVD deaths0.920.790.800.69
p 0.730.340.390.15
Total deaths1.060.961.0990.93
p 0.670.770.370.64
  • Patients known to have hypertension are excluded.

Figure 4.

Survival (cardiovascular mortality) from first screening reading, by diastolic phase‐IV blood pressure (DBP4) divided into five bands and adjusted in a Cox regression for age, body mass index, systolic BP and smoking status (excluding patients previously known to have hypertension). a Females. b Males.

All‐cause mortality

There was a significant interaction between sex and DBP4 band (88–89 vs. 90–99 mmHg, p=0.04). Analysing the sexes separately for all‐cause mortality revealed a significant excess mortality of females with screening band of 88–89mm over 90–99 mmHg (RR 1.56, 95%CI 1.06–2.29, p=0.023). This was not true for males (RR 0.93, 95%CI 0.68–1.27, p=0.64) (Table 3).


This is the first study to suggest that the observers, mainly nurses, recorded a lower reading than they first observed and/or repeated the measurement until they had a value that was more acceptable, with adverse consequences for some female patients. If the DBP4 was 90 mmHg, the patient would be considered for treatment. If the observer thought that they did not need treatment, the pressure appears likely to have been marked down to 88 mmHg. There was no preference for 88 when there was no possible effect on the patient's treatment, for example in measuring DBP5 or systolic pressure, or in patients already known to have hypertension. However, there was a marked preference for 88 mmHg for both males and females in the first, second and third diastolic measurements.

An analysis of pulse pressure showed no step discontinuity at DBP4 readings of 88–90 mmHg, and it is therefore likely that number preference involves only small shifts in blood pressure, e.g. a decision as to whether the DBP4 is 90 or 88 mmHg. This natural process would reassure patients and observers alike. However, the blood pressure reduction to <90 mmHg would be expected to deprive some patients of treatment, and some women with a pressure of 88 mmHg fared badly. In both sexes, the excess of subjects with readings of 88 was about 25%. Who were these subjects? They were both men and women but, in men, those with a DBP4 of 90 were noted to be older than expected. Possibly younger men and both young and older women were consigned to ‘no treatment’. The young men suffered no ill consequences (at least over the mean of 17 years follow‐up). The older women, however, did badly when untreated. This gender difference may be attributable to either a generally higher effort to treat high BP in men at a later date, or to a selectivity in screening that results in depriving high‐risk women of treatment. Our data do not allow us to resolve this issue but the former hypothesis seems more likely.

Study limitations

We acknowledge that the study design allowed us to investigate association but not causality between number preference and mortality, and we draw attention to two particular issues which limit interpretation of the data. Firstly, information was not available on the anti‐hypertensive treatment received by the participants registered for mortality follow‐up. Inclusion of this variable in the survival model would have allowed us to further analyse the sex difference in survival rates that we have demonstrated. For example, if men initially misclassified as ‘88 mmHg’ were later treated earlier and/or more successfully than women, this might explain the subsequent survival rates. Secondly, the use of just one BP reading in initial screening does not conform with current practice, and dilutes the study results.

Other investigators have agreed that terminal digital preference can greatly influence the numbers considered for treatment. For example, Wen and colleagues10 pointed out that in pregnant women changing the definition of hypertension from a systolic pressure ⩾140 mmHg to >140 mmHg halves the prevalence of hypertension from 26% to 13%.

Quality control

The existence of an excess of zero preference suggests furthermore that the random zero sphygmomanometers distributed to participating General Practices as per the study protocol were either not used correctly or not used consistently.

The experience of the GPHSG suggests that research centres (and individual researchers) should be monitored for data quality, particularly where BP measurements are involved. Although this study included meetings of the principal investigators and their staff, it did not specify any formal monitoring of the data during the active phase of the study, and we suggest that such monitoring should be part of all clinical research involving blood pressure measurements. Patterson6 reported in two large teaching practices that practitioners tended to read low at threshold BPs critical for management decisions on managing hypertension. Further research is merited to establish to what extent number preference occurs generally in clinical practice in primary and secondary care. However, since number preference in clinical practice could potentially lead to misclassification of large numbers of patients with adverse clinical consequences, we propose that BP readings should be monitored for number preference and terminal digit preference as part of clinical quality control (Clinical Governance in the UK11).


We wish to acknowledge the work of the following doctors whose practices formed the GP Hypertension Study Group: I. Bradley (Pontesbury), G. Dyker (East Kilbride), P. Ganvir (Manchester), N. Gostick (Rugby), S.R. Mayhew (East Socon), F. Shepherd (Southall), F. Wells (Ipswich), L. Williams (Darlington). We also thank their nursing colleagues who performed much of the clinical work of the study, the patients who participated and Ciba‐Geigy for administering the initial screening program. We acknowledge the support of staff of the department of Medicine for the Elderly at the Hammersmith Hospital and department of Statistics of Imperial College School of Medicine. DW was funded to conduct this study by the NHS Executive London (NHSE‐LRO), Research & Development Responsive Funding Programme. The views and opinions expressed do not necessarily reflect those of the NHSE (LRO) or the Department of Health.


  • Address correspondence to Dr D. Wingfield, Brook Green Medical Centre, Bute Gardens, London W6 7EG. e‐mail: david{at}hammerpcg.org.uk


View Abstract