Intended for healthcare professionals

Primary Care

Prescribing safety features of general practice computer systems: evaluation using simulated test cases

BMJ 2004; 328 doi: (Published 13 May 2004) Cite this as: BMJ 2004;328:1171
  1. Bernard Fernando, general practitioner1,
  2. Boki S P Savelyich, research associate2,
  3. Anthony J Avery (tony.avery{at}, professor of primary health care2,
  4. Aziz Sheikh, professor of primary care research &development3,
  5. Mike Bainbridge, technical director4,
  6. Pete Horsfield, clinical director4,
  7. Sheila Teasdale, service director4
  1. 1Thames Avenue Surgery, Rainham, Kent
  2. 2Division of Primary Care, University of Nottingham, Nottingham NG7 2RD
  3. 3Division of Community Health Sciences: GP Section, University of Edinburgh
  4. 4PRIMIS, University of Nottingham
  1. Correspondence to: Anthony J Avery
  • Accepted 25 March 2004


Over 90% of general practices in the United Kingdom regularly use computers for clinical care.1 These computing systems contain drug interaction alerts, and these are considered useful by most general practitioners.2 Relatively little attention, however, has been paid to other potential safety features for prescribing, such as contraindication alerts. The NHS Information Authority has regulated use of general practice computer systems through a set of rules known as requirements for accreditation, but these contain only general references to safety and there is evidence that they do not prevent contraindicated prescribing.3

Reducing the risks of iatrogenic harm is an important issue for the NHS, and interest has focused on safer prescribing in primary care.4 We undertook a laboratory based evaluation of safety features for prescribing of the four main computing systems used in UK primary care.5

Method and results

We used a two round Delphi approach to reach agreement on the most important safety features of general practice computer systems.5 This involved electronically circulating a list of 55 theoretically derived statements related to safety to 22 members of a selected multidisciplinary expert panel. Statements related to eight broad themes covering key areas in the medicines management process: prescriber alerts, reports and clinical audit, user interface, repeat prescribing, decision support, coding, monitoring, and links to laboratories.

Over 90% of the panel judged 32 of these statements to be important, and these were then used to develop 18 scenarios, which were tested using dummy patient records on the four computing systems. The systems (labelled A, B, C, and D in order to preserve suppliers' anonymity) were independently evaluated at Primary Care Information Services (PRIMIS) laboratories by two members of the project team. To minimise risk of bias, systems were tested with each of the scenarios in random order and data were recorded on to piloted data extraction sheets.

We defined the standards against which the computing systems were to be evaluated. These included appropriate alerts when contraindicated drugs or hazardous drug-drug combinations were prescribed. For each scenario, the safety profile of the computing system was categorised as appropriate or inappropriate. Evaluators compared findings, and an agreed mechanism was available for resolving disagreements should these arise. Finally, to ensure that there were no technical set-up problems that could have accounted for the observed failures, we reported the problems that were identified to the manufacturers and invited comment.

None of the systems produced alerts for all of the 18 scenarios (table). In terms of prescription of drugs with similar names, none of the systems warned for all 10 drug pairs considered.

Responses of computer systems tested for prescribing scenarios

View this table:

The evaluators produced no discrepancies in assessing the safety of systems. Each of the four system suppliers agreed with our assessments.


The safety features of computing systems currently in use in about three quarters of UK general practices have clinically important deficiencies. All may fail to warn in a situation when a warning is expected, thus potentially creating a health hazard to patients.

One solution to this problem is to have more explicit regulations about the situations in which suppliers should implement specific alerts. Because information technology, pharmacology, and medicine are dynamic fields, suppliers of systems and drug databases would need to have regular dialogue with end users about ways of further improving the safety features of these important aides to clinical management. Our discussions with manufacturers indicate that many of the problems uncovered could be resolved, and this work is now being taken forward by the National Patient Safety Agency.


We thank the PRIMIS staff for help with the system set up for the evaluation, GP computer system suppliers for responding constructively to our work, and Judy Cantrill and Caroline Morris for their involvement in the Delphi exercise. This work formed a part of BF's MSc thesis which was co-supervised by Dipak Kalra of CHIME, University College, London.


  • Contributors AA and BF conceived the study, all authors were involved in study design, BF and BS tested the computer systems, all authors were involved in the analysis, and BF, BS, AS, and AA wrote the paper. AA is guarantor.

  • Funding National Patient Safety Agency.

  • Competing interests Until October 2002 MAB was paid as a medical consultant to Torex Health.

  • Ethical approval Not required.