OUP user menu

Cluster‐randomized, controlled trial of computer‐based decision support for selecting long‐term anti‐thrombotic therapy after acute ischaemic stroke

DOI: http://dx.doi.org/10.1093/qjmed/hcg019 143-153 First published online: 1 February 2003


Background: Identifying the appropriate long‐term anti‐thrombotic therapy following acute ischaemic stroke is a challenging area in which computer‐based decision support may provide assistance.

Aim: To evaluate the influence on prescribing practice of a computer‐based decision support system (CDSS) that provided patient‐specific estimates of the expected ischaemic and haemorrhagic vascular event rates under each potential anti‐thrombotic therapy.

Design: Cluster‐randomized controlled trial.

Methods: We recruited patients who presented for a first investigation of ischaemic stroke or TIA symptoms, excluding those with a poor prognosis or major contraindication to anticoagulation. After observation of routine prescribing practice (6 months) in each hospital, centres were randomized for 6 months to either control (routine practice observed) or intervention (practice observed while the CDSS provided patient‐specific information). We compared, between control and intervention centres, the risk reduction (estimated by the CDSS) in ischaemic and haemorrhagic vascular events achieved by long‐term anti‐thrombotic therapy, and the proportions of subjects prescribed the optimal therapy identified by the CDSS.

Results: Sixteen hospitals recruited 1952 subjects. When the CDSS provided information, the mean relative risk reduction attained by prescribing increased by 2.7 percentage units (95%CI −0.3 to 5.7) and the odds ratio for the optimal therapy being prescribed was 1.32 (0.83 to 1.80). Some 55% (5/9) of clinicians believed the CDSS had influenced their prescribing.

Conclusions: Cluster‐randomized trials provide excellent frameworks for evaluating novel clinical management methods. Our CDSS was feasible to implement and acceptable to clinicians, but did not substantially influence prescribing practice for anti‐thrombotic drugs after acute ischaemic stroke.


A key challenge in the clinical management of acute ischaemic stroke and transient ischaemic attack (TIA) is to select the optimal strategy to reduce the risk of further cerebrovascular events. Among the available therapeutic and surgical options are the long‐term use of antiplatelet therapy (aspirin, dipyridamole and clopidogrel, alone or in combination) and anticoagulation with warfarin.1 Many large clinical trials have demonstrated the efficacy of these agents. Moreover, meta‐analyses2–,5 give us high‐level evidence of efficacy in the general post‐ischaemic stroke population. It can be difficult to translate these group‐level findings into a risk‐benefit ratio for the clinical management of an individual patient, as comorbid disease and relative contra‐indications to therapy must also be taken into account. The patients with greatest potential to benefit from anti‐thrombotic therapy are also often those at highest risk of adverse events on treatment. Use of a computer‐based decision support system (CDSS) could assist in the selection of the most appropriate anti‐thrombotic therapy.

A CDSS is ‘software designed to directly aid in clinical decision making about individual patients’.6 Previous CDSS implementations have involved provision of prognostic information to assist clinical management, issuing of reminders to general practitioners regarding appropriate preventive strategies, assistance with differential diagnosis and selection of appropriate drug treatments. Testing of computer‐based decision support systems has become more common in recent years with the increased availability of computers in the clinical setting. A systematic review covering the years 1974–1998 found 68 trials which considered the effect of a CDSS on clinician performance or patient outcomes.7 In 65 studies assessing physician performance, 66% found a beneficial effect, while patient outcome was improved in 43% of the 14 trials in which this was measured.

This study considers prescribing practice for antiplatelet and anticoagulant drugs following acute ischaemic stroke or TIA. Anti‐thrombotic therapy reduces long‐term incidence of ischaemic events at the expense of an increase in haemorrhagic event rates. The CDSS which we evaluated estimates, for the individual acute stroke or TIA patient, the annual risks of recurrent ischaemic stroke, haemorrhagic stroke, myocardial infarction, other ischaemic vascular events and other haemorrhagic complications associated with each possible antiplatelet or anticoagulant therapy. The information provided by the CDSS enables an informed decision on prescribing to be made.

Cluster‐randomized trials provide an excellent structure within which novel approaches to clinical management may be evaluated. Our study aimed to quantify the impact on prescribing practice of providing patient‐specific information from the CDSS.


CDSS intervention

The CDSS in this study was an expert system.8 Expert systems have been used in several medical contexts, including electromyography,9 stroke diagnosis10 and stroke localization.11 The knowledge base of our CDSS incorporates published efficacy data from all relevant randomized controlled trials and meta‐analyses. Risk factor prevalences were obtained from acute stroke unit databases. Published results from observational studies provided information on haemorrhagic event rates during antiplatelet and anticoagulant therapy. Where high‐level evidence was unavailable from randomized trials or observational studies, information from an expert panel of five experienced stroke clinicians was used for the knowledge base. This assisted with risk estimation in clinical subgroups for which there was no trial evidence. Information about the CDSS knowledge base was available on request to all physicians participating in PRISM.

The CDSS uses aspects of the medical history of the patient and clinical findings (Table 2) to estimate the annual risk of each of five categories of outcome event: recurrent ischaemic stroke, primary intracerebral haemorrhage, myocardial infarction, other ischaemic complications and other haemorrhagic complications. The rates include fatal and non‐fatal events. The conditions included in each event category are listed in the appendix with their World Health Organization ICD9 codes.12 The CDSS estimates the event risks associated with six potential therapeutic strategies: (1) long‐term anticoagulation with warfarin; (2) antiplatelet therapy with aspirin; (3) antiplatelet therapy with dipyridamole; (4) aspirin and dipyridamole in combination; (5) warfarin and aspirin in combination; or (6) no antiplatelet or anticoagulant therapy. For each potential therapy, the CDSS summarized the estimated event rates to clinicians in a graph of total ischaemic event risk and total haemorrhagic event risk. The CDSS calculates event rates, but does not incorporate information on treatment costs.

During the course of this trial, the antiplatelet agent clopidogrel became licensed. Since it would have been inappropriate to alter the CDSS knowledge base during the study, the system did not include data on clopidogrel. Such information could be incorporated in a future version of the system. The software HUGIN13 performed the CDSS calculations.

Study population

In‐patients or out‐patients with a clinical diagnosis of acute ischaemic stroke or TIA were eligible for the study if this was a first investigation of an event occurring within the preceding four months. We used the World Health Organization definitions of stroke and TIA.14

Table 1 summarizes the study entry criteria. The CDSS takes account of the presence of hypertension, peptic ulcer disease, a history of accidental falls and expected poor compliance of the patient with therapy. Therefore, these relative contra‐indications to anticoagulation did not require exclusion.

View this table:
Table 1 

PRISM inclusion and exclusion criteria

View this table:
Table 2 

Patient characteristics at entry to study

Study design

We used a cluster‐randomized, controlled before‐and‐after design. We did not randomize at patient or doctor level, since knowledge gained from the CDSS might then have been applied to patients randomized to the control group, leading to underestimation of any effect of the CDSS.6 Hospitals were randomized as either control or intervention centres, stratified before randomization according to the date on which patient recruitment was due to start. The stratification ensured that the treatment periods would be comparable between the CDSS and control groups. Random number tables were used to allocate one hospital from each pair to the CDSS group. Randomization was performed by a member of the study group (CJW) not directly involved in enrolment of patients to the trial.

During phase one, routine prescribing practice was observed and recorded in intervention and control centres. This data collection continued in control centres in the second phase, while in intervention centres, data were recorded after the CDSS had provided clinicians with patient‐specific estimated event rates for each possible antiplatelet or anticoagulant therapy. This design enabled any effect of the CDSS to be distinguished from other changes in prescribing practice over time, since changes over time in intervention centres could be compared to those in control centres. Each study phase was planned to last for 6 months, subject to acceptable recruitment rates. Data collection and data entry procedures were identical for control and intervention centres.

Medical staff could not be blinded to the status of their hospital (control or intervention) since information was being provided. Patient consent was not required, as the study simply observed prescribing practice. During phase two, patients in intervention centres were informed that the study was being conducted and were given an information sheet regarding the study.

The multi‐centre research ethics committee (Scotland) approved the protocol. Two local research ethics committees amended the protocol to require informed patient consent from all patients included in the study; a third required informed consent from patients managed with decision support from the CDSS. In general, the issue of patient consent in cluster randomized trials remains unresolved.15

Method of presenting CDSS results

Baseline clinical data (Table 2) were entered via an automated telephone data entry system as soon as possible after the hospital admission (in‐patients) or clinic appointment (out‐patients) of a study patient. The CDSS processed these data to estimate the event rates for each category of event and each possible therapy. During phase two, intervention centres received, immediately following data entry, a return facsimile containing a graph of the CDSS output. This patient‐specific information was inserted in the case record and was available to medical staff to support the decision on the most suitable anti‐thrombotic therapy for that individual. Finally, the actual long‐term anti‐thrombotic therapy prescribed for each patient was recorded.

Primary and secondary outcome measures

For each patient, the CDSS was used to calculate the relative risk reduction (RRR) in ischaemic and haemorrhagic vascular events which was achieved by the actual therapy prescribed versus the option of ‘no antiplatelet or anticoagulant therapy’. The primary endpoint compared, between control and intervention centres, the changes from phase one to phase two in the mean RRR achieved. This endpoint indicated how closely the prescribing decision had followed the advice from the CDSS.

Secondary outcome measures were: (1) comparison of the proportion of ‘optimal’ treatments (the treatment that would give the lowest estimated event rates according to the CDSS) for patients managed with and without the CDSS; (2) comparison of the proportions of patients receiving each anticoagulant or antiplatelet therapy. We also identified factors which resulted in anticoagulation being withheld.

Sample size and statistical analysis

The sample size must be inflated for a cluster‐randomized trial, due to the intra‐class correlation among prescriptions to patients within each hospital.16 Pilot study data of routine prescribing practice indicated a mean total RRR of 18%. To detect an absolute change of 5% in this value, a single‐phase study would require 27 centres to recruit 40 patients each, a total of 1080 patients (power 80%, two‐sided significance level 5%). This assumes an intra‐class correlation of 0.15, a typical figure for such studies. Our study design gains power, since the additional ‘observation only’ phase allows adjustment for baseline data in each centre. The sample size was calculated using the SSC software (Health Services Research Unit, University of Aberdeen).

Baseline patient characteristics were compared between intervention and control groups and between study phases using the Mann‐Whitney test (continuous variables) and the χ2 test (categorical variables). Primary and secondary endpoints were analysed according to the intention‐to‐treat principle by multilevel modelling17 using the MLwiN software.18 This accounts for the intra‐class correlation within hospitals, and allows modelling of the mean and variability of data. The primary endpoint was assessed in two ways. Initially, study phase, centre type (intervention or control) and patient type (in‐patient or out‐patient) were included as potential factors in the multilevel model in addition to correcting for the estimated event rate of the patient if no therapy were prescribed and testing for the CDSS effect. In the second analysis, any of the patient‐specific clinical features from the baseline data collection were also available for inclusion in the model.

The CDSS did not calculate event rates for clopidogrel prescription, since clopidogrel was not licensed at the beginning of our study and it would have been inappropriate to alter the CDSS during the trial. Similarly, CDSS event rate calculations were not available for patients prescribed unusual therapies (for example, warfarin and dipyridamole in combination) not incorporated in the CDSS knowledge base. These subjects were included in the analysis with their event rate being assumed to be the median of those calculated by the CDSS for all possible therapies. By comparing the findings using different risk assignment methods, we assessed the sensitivity of the results to this method.

Clinicians in intervention centres were surveyed in order to elicit their reaction to the implementation of the CDSS and the information it provided.


Nineteen centres obtained local research ethics committee approval: of these, 16 actually participated. Figure 1 illustrates the flow of participants through the study. We included 1147 individuals in control centres as in‐patients (86%), the remainder being out‐patients. The corresponding figure for intervention centres was 491 (81%). Diagnosis was confirmed by CT or MR scan in 1534 patients (79%).

Table 2 shows the characteristics of patients at entry to the trial for intervention and control centres within each study phase. In general, the study groups were well balanced, although vascular risk factors were slightly more prevalent in patients from intervention centres.

Table 3 summarizes the primary and secondary endpoint results. The intra‐class correlation coefficient of the estimated RRR among patients within hospitals was 0.052. Table 4 gives the results of the multilevel modelling for the primary endpoint. Providing information from the CDSS did not substantially alter the mean estimated RRR achieved by prescribing antiplatelet or anticoagulant therapy. Similarly, supplying information from the CDSS did not influence the proportion of patients receiving the optimal therapy identified by the CDSS. (Table 5)

Certain clinical features (atrial fibrillation (AF), previous myocardial infarction, active peptic ulcer, carotid artery disease) were statistically significantly related to the RRR that was achieved. However, the estimate of the influence of the CDSS on mean RRR or whether optimal therapy was prescribed remained largely unchanged if these variables were included in the relevant multilevel model.

At baseline, prescribing was similar in control and intervention centres, although control centres had a lower rate of aspirin prescription, and prescribed aspirin and dipyridamole in combination more often. Aspirin monotherapy decreased over time, while clopidogrel use and anticoagulation increased in frequency. Anticoagulation, use of clopidogrel and of aspirin and dipyridamole in combination increased more in intervention centres than in control centres during the period when the CDSS data were provided.

The primary and secondary endpoint results were insensitive to the choice of risk assignment to subjects prescribed clopidogrel and other therapies not included in the CDSS knowledge base. If, rather than the median, these patients were assigned the lowest of the risks calculated, or the greatest of the risks, the results were largely unchanged.

Long‐term therapy was recorded in 269 of the 271 patients with AF who survived to discharge. The effect of the CDSS in this important clinical subgroup did not differ from that in the whole study group. One hundred and twenty‐two patients (45%) with AF were anticoagulated long‐term with warfarin alone or in combination with an antiplatelet agent. Clinical features that influenced the proportion of patients anticoagulated were: whether poor compliance with therapy was expected (poor compliance expected, 14% anticoagulated vs. poor compliance not expected, 48% anticoagulated, p<0.01); and whether there was a history of accidental falls (history of falls, 16% anticoagulated vs. no history of falls, 46% anticoagulated, p=0.02). Most AF patients not anticoagulated were prescribed aspirin alone (38%) or in combination with dipyridamole (10%). Twelve patients with AF (4%) were prescribed neither anticoagulant nor antiplatelet therapy. Among the reasons given were contraindication to anticoagulation (4 patients), gastrointestinal or other haemorrhagic problem (3), expected poor compliance (2) and that the patient was aged >90 years (3). The remaining AF patients were prescribed clopidogrel (3 patients) or dipyridamole (3 patients).

Nine clinicians (at least one from each CDSS centre) responded to the survey. All respondents confirmed that the CDSS information was available sufficiently soon to be of use in the prescribing decision. The format in which the evidence was presented was acceptable to eight clinicians. Three respondents disagreed with the CDSS (for example, when lower risks were estimated for aspirin and dipyridamole in combination vs. aspirin, and when anticoagulation was recommended for patients in whom it was considered unsuitable). Four centres had local anti‐thrombotic prescribing guidelines or recommendations in place. Finally, 55% (5/9) of respondents felt that the CDSS had influenced their prescribing practice.

Figure 1. 

Participant flow through the trial.

View this table:
Table 3 

Summary of endpoints

View this table:
Table 4 

Relative risk reduction: multilevel model

View this table:
Table 5 

Optimal therapy prescription: multilevel model


Summary of results

Trends in our data suggest some positive effect of the CDSS on prescribing practice. Reduced reliance on aspirin monotherapy in intervention centres corresponded to an increase in anticoagulation and the combination of aspirin with dipyridamole. We have, however, ruled out a large influence of the CDSS: an absolute improvement in mean relative risk reduction of greater than 5.7 percentage units has been excluded. The proportion of patients being prescribed the optimal therapy identified by the CDSS increased from phase I to phase II in control and, more markedly, in intervention centres. This difference between control and intervention centres was not statistically significant. Overall, 69% of patients did not receive the optimal therapy identified by the CDSS (Table 3).

In addition to exploring changes in the mean value of primary and secondary endpoints, multilevel modelling enables us to investigate whether the variability in study endpoints was affected by the CDSS. This is of interest, since it is conceivable that the mean value may remain unchanged following the introduction of the CDSS, while excess variation in prescribing present at baseline may be suppressed. Variability in the primary endpoint decreased in all study centres from phase I to phase II; the decrease was greater in centres in which the CDSS was implemented (a statistically non‐significant decrease of 18% vs. 5%).

Internal and external validity

Characteristics of patients from the CDSS and control centres at baseline were comparable, contributing to the internal validity of the study. The statistical analysis also allowed correction for any differences in clinical features that were present. The study design enabled any effect of introducing the CDSS to be differentiated from trends in prescribing practice over time. Aside from the introduction of the CDSS, patients were managed according to routine practice in all study centres during the active phase.

Study centres provided stroke care in a variety of environments. Seven centres were general medical wards, five incorporated an acute stroke unit or clinic, three were geriatric wards and one was a neurology ward. The distribution of care provision types was similar for intervention and control centres. Clinical guidelines on prescribing of anti‐thrombotic therapy were implemented in three control centres and four intervention centres. Five control centres included out‐patients in the study compared to three CDSS centres. The broad inclusion criteria for PRISM meant that we studied a wide spectrum of the patients for whom anti‐thrombotic therapy must be selected after stroke or TIA.

A study such as this may tend to include centres with specialist knowledge of, or an interest in, stroke care. We observed little evidence of a difference in the effect of the CDSS across the various care provision structures and between specialist and generalist centres. However, a greater influence of the CDSS may have been demonstrable in unselected centres.

Interpretation of findings

Several factors may have influenced the effect of the CDSS on prescribing. Structured data collection of the patient case history in all study groups may itself influence patient care.19 We corrected for this ‘checklist effect’ by including the first phase where clinical data were recorded in all centres. Presenting treatment benefit as a relative, rather than absolute, risk reduction leads to a greater impact of the evidence.20 In this study we effectively presented absolute risks, to describe separately the risk of haemorrhagic and ischaemic events. In general, clinicians in CDSS centres were favourably disposed towards the information provided by the CDSS; however, in four of these centres, local prescribing guidelines were also in place, and may have influenced prescribing practice to a greater extent than the CDSS. The guidelines and the CDSS information would not necessarily have been in agreement for all patients, as guidelines may take account of treatment costs and do not provide patient‐specific decision support.

Information from the CDSS may also be used to engage the patient in discussion of the risks and benefits of treatment—particularly regarding anticoagulation with warfarin. Such clinical management decisions often invoke patient perceptions of risk, and using the CDSS to present the estimated event rates in a graphical form is one method of eliciting patient opinion.

Notably, there was considerable variation among centres in prescribing practice for anti‐thrombotic therapy. Several factors may contribute, including the case mix in each centre, differences among local guidelines on prescribing, availability of the more costly antiplatelet alternatives to aspirin and differing interpretations of the results of trials of aspirin and dipyridamole in combination21 and clopidogrel.22 There is still considerable uncertainty regarding the optimal therapy for patients experiencing recurrent events while on aspirin therapy. The absence of a large effect of the CDSS on practice is therefore unlikely to be due to uniform best practice already being present in all study centres at baseline.

Of the patients who were not prescribed the optimal therapy identified by the CDSS, 795 (74%) were prescribed aspirin when the CDSS recommended aspirin and dipyridamole in combination. This may reflect uncertainty among clinicians over the interpretation of trials such as ESPS‐2.21 Alternatively, their prescribing decisions may have taken account of cost‐effectiveness if there was only a small difference in estimated event rates in favour of the more expensive therapy. If the weight of evidence from clinical trials was perceived to be inadequate by medical staff, this may have limited the influence of the CDSS on prescribing. The potential for benefit from a CDSS will increase as the body of evidence from trials accumulates over time.

Although the effects of introducing the CDSS in this secondary care setting were limited, it may have greater utility in primary care, where clinicians have less experience in the decision‐making regarding anti‐thrombotic therapy following ischaemic stroke or TIA. Although long‐term preventive therapy is generally commenced in secondary care, intercurrent events often change the clinical scenario. For example, if instability and an increased risk of falls develop, or if gastrointestinal complications take place, the family doctor must then review the therapeutic options. The vascular risk screening and clinic programme currently being developed in UK primary care may provide a suitable context for implementation of a CDSS.

Cluster‐randomized trials

The increasing use of cluster‐randomization in trial design reflects its suitability in addressing questions regarding clinical management. As outlined in the Methods section, cluster‐randomisation guards against biases such as ‘contamination’: the application by clinicians of knowledge gained from the CDSS to patients randomized to be managed without the CDSS.

Cluster‐randomized trials have considerable ethical and practical advantages for the evaluation of rehabilitation strategies in acute stroke. For example, consider a trial assessing the impact of specialist neurophysiological training of physiotherapists in order to promote the recovery of postural control in stroke patients. If a physiotherapist had been given such specialist training, it would be a good use of resources to allow all patients of that physiotherapist to be eligible for specialist therapy. Randomization of physiotherapists (prior to their allocation to specialist training) rather than patients would enable this. The benefits of cluster‐randomization do have a cost, however: the sample size required is substantially larger than for an equivalent trial randomized at patient level.16

Further research and conclusions

The outcome measures in this study assessed clinical practice rather than patient outcomes. If a substantial effect of the CDSS on clinical management had been observed, the next logical step would have been to conduct an equivalent study which measured actual vascular event rates. Such a study would require a much larger number of patients. A subset of the PRISM cohort will be followed up, via Scottish databases of routinely recorded hospital data, to investigate the relationship between actual event rates in the general ischaemic stroke population and predictions from the CDSS. These analyses will inform future developments of CDSSs in cerebrovascular medicine and the designs of CDSS trials that measure patient outcomes.

In summary, we have shown that providing clinicians with information from a CDSS is technically achievable but did not substantially influence prescribing practice of anti‐thrombotic therapy following acute ischaemic stroke or TIA. The heterogeneity in prescribing practice we observed among hospitals suggests that consensus on best practice has yet to be adopted. As evidence from clinical trials on efficacy continues to accumulate rapidly, clinical guidelines and decision support are likely to play a role in optimizing prescribing practice.

Appendix: Outcome category events and corresponding ICD9 codes

Ischaemic stroke or TIA

433 Occlusion and stenosis of precerebral arteries

434 Cerebral thrombosis or embolism

435 Transient cerebral ischemia

436 Acute cerebrovascular disease

Haemorrhagic stroke

430 Subarachnoid haemorrhage

431 Intracerebral haemorrhage

432 Unspecified intracranial haemorrhage

Myocardial infarction

410 Acute myocardial infarction

Other ischaemic complications

246.3 Infarction of thyroid

253.8 Pituitary gland infarction

255.4 Adrenal gland infarction

289.5 Infarction of spleen

415.1 Pulmonary embolism or infarction

444.8 Embolism or thrombosis of other specified artery

444.9 Embolism or thrombosis of other unspecified artery

453.9 Venous embolism or thrombosis of unspecified site

557.0 Infarction or embolism of bowel or intestine; mesenteric infarction

573.4 Hepatic infarction

593.8(1) Renal (artery) embolism or infarction

602.8 Infarction of prostate

611.8 Breast infarction

620.8 Infarction of ovary or fallopian tube

Other haemorrhagic complications

246.3 Haemorrhage of thyroid

252.8 Haemorrhage of parathyroid gland

255.4 Adrenal gland haemorrhage

280 Iron deficiency anaemias

285.1 Acute post‐haemorrhagic anaemia

289.5(9) Haemorrhage of spleen

360.4(3) Haemophthalmos

362.8(1) Retinal haemorrhage

363.6 Choroidal haemorrhage

364.4(1) Haemorrhage of iris or ciliary body

372.7(2) Conjunctival/subconjunctival haemorrhage

379.2(3) Vitreous haemorrhage

379.8 Other haemorrhagic eye and adnexa disorders

386.8 Labyrinthine haemorrhagic disorders

423.0 Haemopericardium

429.8 Other ill‐defined heart diseases

448.9 Capillary haemorrhage

459.0 Unspecified haemorrhage

478.2(0) Haemorrhage of pharynx

511.8 Haemothorax

523.8 Gingival or periodontal haemorrhage

528.9 Oral soft tissue haemorrhage

529.9 Haemorrhage of tongue

530.8(2) Oesophageal haemorrhage

533.0 Acute peptic ulcer with haemorrhage

533.2 Acute peptic ulcer with haemorrhage and perforation

537.8 Haemorrhage of stomach or duodenum

568.8(1) Haemoperitoneum

569.3 Haemorrhage of rectum or anus

573.8 Haemorrhage of liver

577.8 Haemorrhage of pancreas

578.9 Gastrointestinal haemorrhage

593.8(1) Renal haemorrhage

596.7 Haemorrhage into bladder wall

596.8 Bladder haemorrhage

599.7 Haematuria

599.8 Urethra or urinary tract haemorrhage

602.1 Prostate haemorrhage

607.8(2) Haemorrhage of corpus cavernosum or penis

608.8(3) Haemorrhage of other male genital organs

623.8 Haemorrhage of vagina

624.5 Haematoma of vulva

627.1 Postmenopausal bleeding

629.8 Haemorrhage of other female genital organs

665.7(0–2,4) Pelvic haematoma

719.1 Haemarthrosis

782.7 Spontaneous petechiae

784.7 Epistaxis

784.8 Haemorrhage from throat

786.3 Pulmonary haemorrhage

862.2 Haemorrhage of other intrathoracic organs


The PRISM Study Group were: Glasgow, UK: Western Infirmary (K.R. Lees, I. Sim, C.J. Weir), Stobhill Hospital (L. Erwin, C. McAlpine, J. Rodger), Southern General Hospital (T. Jones, K.W. Muir); Dundee, UK: Ninewells Hospital (H.W. Fraser, R.S. MacWalter); Paisley, UK: Royal Alexandra Hospital (A. Dorward, G. Gorman, A. Hussein, L. Quate); Ayr, UK: Ayr Hospital (S. Ghosh, K. Musbahi, U. Rutherford); Leicester, UK: Leicester General Hospital (W. Brooks, J. Hamilton, T.G. Robinson); London, UK: Greenwich District Hospital (M.S. Ali, L. Gogola), Whittington Hospital (L. Al‐Dhahir, S. Gerrie, G.S. Rai); Ormskirk, UK: Ormskirk District General Hospital (J. Horsley); Macclesfield, UK: Macclesfield District General Hospital (M. Salehin, D.J. Walker); Airdrie, UK: Monklands Hospital (D.G. Grosset); Brecon, UK: Bronllys and Llandrindod Wells Hospitals (A. Dunn, S. Manthri, G. Woodman); Magdeburg, Germany: Otto‐von‐Guericke‐Universität (M. Goertler, T. Treuheit, C.‐W. Wallesch); Wolverhampton, UK: New Cross Hospital (A. Brown, D.F. D'Costa, E.V. McLelland, J. Sinclair); Carluke, UK: Law Hospital (K. Boyle, A. Hendry).

The writing committee were: C.J. Weir, K.R. Lees, R.S. MacWalter, K.W. Muir, C.‐W. Wallesch, E.V. McLelland, and A. Hendry.

This study was funded as part of a Medical Research Council Special Training Fellowship in Health Services Research. Professor Jeremy Grimshaw (Health Services Research Unit, University of Aberdeen) advised on the study design and sample size calculations. The investigators of the Stroke Prevention in Atrial Fibrillation (SPAF) clinical trial group assisted in the CDSS validation by making data from the SPAF trials available for subgroup analysis. Min Yang (Institute of Education, London) advised on the structure of the multilevel statistical models in the data analysis.


  • Address correspondence to Dr C.J. Weir, Department of Medicine & Therapeutics, University of Glasgow, Gardiner Institute, Western Infirmary, Glasgow G11 6NT. e‐mail: cjw2f{at}clinmed.gla.ac.uk

  • *See Acknowledgements for a list of members.


View Abstract