A computer based clinical decision support system was written for the two most commonly used practice computing systems (EMIS and AAH Meditel) so that it could be incorporated into routine clinical care. The system is identical to the New Zealand guidelines for the management of hypertension,2
except that absolute risk is presented numerically rather than pictorially. The following patient information is required to ascertain absolute cardiovascular risk: sex, age, diabetes, smoking, blood pressure, cholesterol, body mass index, symptomatic cardiovascular disease, family history of ischaemic heart disease, and familial hypercholesterolaemia. The system then calculates the patient's five year risk of a fatal or non-fatal cardiovascular event (newly diagnosed angina, myocardial infarction, coronary heart disease, stroke, or transient ischaemic attack).
We invited all 96 practices in Avon using the EMIS and AAH Meditel computing systems to participate in the study. Practices agreeing to participate in the study were firstly stratified by computer system and then assigned by simple random allocation to use the computer based clinical decision support system and a cardiovascular risk chart (which gives identical information about risk), the risk chart alone, or usual care (no information given about cardiovascular risk). Randomisation was performed with a table of random numbers by a researcher not involved in the study and who was blind to the identity of the practices.
All patients aged 60-80 years with a diagnosis of hypertension and a record of having been prescribed antihypertensive drugs in the previous year were eligible. Thirty eligible patients were randomly sampled from each practice list by using either the computer system's built-in sampling facility (EMIS practices) or a random sampling program on a personal computer (AAH Meditel practices).
The main study was performed by general practitioners (n=74) and practice nurses (n=11), depending on procedures for management of hypertensive patients within each practice. The general practitioners and nurses were trained to use the computer based clinical decision support system by one of us (AM). Each patient's blood pressure was measured on the day of attendance. Other risk factors were extracted from the patient's notes and cardiovascular risk recorded either automatically (in computer practices) or manually by the general practitioner or nurse (in chart practices). Data were missing only for total and high density lipoprotein cholesterol concentrations. The computer system assumed missing data to have a value representing the lowest addition to absolute risk, and general practitioners and nurses in the chart only practices were instructed to do likewise. A sensitivity analysis was performed for the outcome of absolute cardiovascular risk in which missing values were assigned the median for each trial group. Follow up was at six and 12 months. Because of the nature of the study, neither the doctors and nurses nor the patients were blind to their study group. The study took place from September 1996 to September 1998.
The primary outcome was the percentage of patients in each group with five year cardiovascular risk
10%. Secondary outcomes were systolic and diastolic blood pressure and prescribing of cardiovascular drugs. Although follow up data were collected at six and 12 months, the primary follow up was at 12 months, and only these results are presented here.
We managed and analysed data using Stata Statistical Software.7
Baseline comparability of the groups was investigated by descriptive statistics. All analyses comparing the groups at follow up were conducted on an intention to treat basis. We used multivariable regression models and adjusted for the value of the outcome variable at baseline and practice computer system. Since randomisation was by practice, we also corrected for clustering using procedures in Stata to derive robust estimates of standard error. Lastly, we tested for an interaction effect of baseline risk level and trial arm.
To assess intensity of treatment, we collected prescribing data for cardiovascular drugs at baseline and six months (there being insufficient time at the end of the trial to repeat the exercise). It was decided in advance to consider this variable in three categories as follows: 0-1; 2; and 3 or more different classes of cardiovascular drugs. The distribution at six months was compared across the three groups by simple χ2 techniques and multinomial logistic regression models that allowed for the corresponding distributions at baseline.
The trial was designed principally to detect a difference between the two intervention arms of the trial, computer based clinical decision support system plus chart versus chart alone. Based on previous work,8
we estimated that 55% of patients would have an absolute cardiovascular risk over five years of
10%. The sample size was designed to detect a difference between these two groups of 20 percentage points in the proportion of patients with five year risk
10%. To allow for randomisation by practice, the sample size was inflated by a factor of two based on an intrapractice correlation coefficient of 0.0551.9,10
With this inflation factor, 80% power and two tailed 5% α, the required sample size was 190 in each of the two intervention groups. Twenty seven practices agreed to participate in the trial: 10 were therefore randomly allocated to each of the two intervention arms and seven to the usual care arm. To ensure sufficient numbers of patients with adequate follow up data, 30 patients were randomly sampled from each practice.