METHODS

Practices

Practices from three PCTs in the north west and east Midlands areas of England were eligible for inclusion in the study if they fulfilled the following inclusion criteria:

• Member of either the Trent Focus Collaborative Research Network or the Primary Care Information Services (PRIMIS) network.

• Running the EMIS 5 byte operating system.

• MIQUEST (Morbidity Information Query and Export Syntax) software installed.

• Electronically laboratory linked.

Inclusion was restricted to members of the specified networks because these organisations place an emphasis on the quality of the electronic patient record data.

Applying the indicators in practice

The number of potential PDRM events (as defined by the indicators) was assessed retrospectively by searching the electronic patient record of all patients aged 18 years and over in each practice over a 2 year 3 month time frame (1 November 1999 to 31 January 2002). The computer searches were run in each practice by a member of the research team (CJM, SR or VSH) between July and November 2002. This was achieved using the MIQUEST computer software program. The steps involved in this process are shown in fig 3.

PDRM indicators

A series of indicators for PDRM were applied in the GP practices. A summary of their derivation is shown in fig 1. From this previous work we had validated 39 PDRM indicators suitable for application in primary care.^7–9 However, only 29 indicators were successfully applied in this study. The reasons for the exclusion of 10 indicators are shown in box 1.

The indicators encompass a variety of different clinical situations and all take the form of an adverse therapeutic outcome (the PDRM event) resulting from an associated pattern of patient care.

Data analysis

Up to six individual queries were required to collect the relevant data for each indicator, thus considerable data manipulation was required after data extraction. This process was automated as far as possible and is shown in fig 4.

Completing this process identified the number of potential PDRM events for each indicator and therefore the most commonly occurring events in the study practices.

The study was approved by the local research ethics committee in each locality and the University ethical committee.

RESULTS

Fifteen practices were approached, 12 of which agreed to participate in the study. However, data from only nine practices were used as two were not laboratory linked and one did not have functioning software. In addition, two of the practices had linked computer support and were therefore treated as a single practice for the purposes of this study.

The number of GP partners ranged from 1 to 7. Five practices held GP training status and they were located in urban (n = 3), suburban (n = 3), and rural areas (n = 2).

PDRM events

A total of 49 658 electronic patient records were identified for patients aged 18 years and over which were eligible for interrogation by the MIQUEST queries. In all, 507 potential PDRM events were identified, giving an overall incidence of PDRM of 1.0%. The incidence of PDRM at the individual practice level is shown in table 1. Table 2 shows the total number of events identified for each indicator and the range of numbers of events for each indicator in individual practices. It can be seen from table 2 that four indicators accounted for the majority of the events (302/507, 59.6%). These related to the use of non-steroidal anti-inflammatory drugs in patients with congestive heart failure or hypertension, lack of monitoring of potassium and creatinine in patients prescribed angiotensin converting enzyme (ACE) inhibitors, and the use of hypnotic-anxiolytic agents. No events were identified for five indicators, 1–10 for 11 indicators; 11–20 for five indicators, and 21–40 for four indicators.

DISCUSSION

This pilot study has shown that a substantial number of potential PDRM events occur in English primary care. The computerised queries were able to interrogate GP systems, resulting in the identification of potential PDRM events in one in 100 electronic patient records.

Identification of potential PDRM events enables us to consider strategies that are likely to have the greatest impact on reduction of future PDRM. This in turn will benefit patients by improving the safety and the quality of health care and also have a positive impact on healthcare resource utilisation.

It is notable that most of the indicators contributed to very few events, while four (the prescribing of NSAIDs and hypnotic-anxiolytics and the monitoring of ACE inhibitors) contributed to approximately 60% of the PDRM events. However, the clinical implications for individual patients may be great, irrespective of the number of events identified by each indicator. To date, limited research has been undertaken on PDRM. As this work took place in the North American healthcare system,^6,12,13 only cautious comparisons can be made. Nevertheless, despite the fact that MacKinnon and Hepler¹² applied PDRM indicators within a managed care organisation using the database of a hospital based healthcare plan in Florida, a similar pattern emerged. A small number of indicators accounted for the majority of the PDRM events. Furthermore, some similarities existed in the clinical issues that contributed to the seven most frequently occurring PDRM events,¹² with NSAIDs, ACE inhibitors, and failing to prescribe a β blocker after myocardial infarction appearing in both data sets.

However, when interpreting the study results a number of factors need to be considered. Although a temporal relationship was identified between drug therapy and the PDRM event, it does not necessarily mean that the drug was the direct or only cause of that event. Furthermore, the quality of our data is dependent upon the quality of the data present on the practice system. Although prescribing data are generally of better quality than diagnostic or lifestyle data on primary care computer systems,¹⁴ we accept that the quality of data recording in the practices was beyond our control. For this reason, the PDRM incidence data should be interpreted cautiously. At present it would be inappropriate to use these data to draw comparisons between practices but, with current moves to improve data recording in the UK and the implications of the new GP contract,¹⁵ computerised data from general practices are likely to be more reliable in the future and appropriate comparisons may then prove useful. Despite some practical difficulties, electronic patient records have considerable potential in the detection of PDRM.

We have shown that a substantial number of potential PDRM events are occurring in primary care. However, because of the problems of assigning direct causality of the drug to the event, rather than aiming to collect prevalence data our ultimate purpose is to use the indicator data to generate discussion in facilitated multidisciplinary discussion forums attended by key practice personnel (GPs, nurses, pharmacists and the practice manager). These will use the key principles of root cause analysis (prioritisation, root cause identification and root cause elimination).¹⁶ The outcome of these meetings will form the focus of a subsequent paper describing how the overall process can facilitate improvements in medicines management and ultimately the quality of patient care.