Article Text

Download PDFPDF

Drug related morbidity
Safety from numbers: identifying drug related morbidity using electronic records in primary care
  1. G Elwyn
  1. Correspondence to:
 Professor G Elwyn
 Primary Care Research Group, University of Wales Swansea Clinical School, Swansea SA2 8PP, UK;

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

The use of electronic clinical data to identify temporal associations between drug prescribing and patient morbidity

The benefits of creating a searchable patient record are slowly emerging, although it is arguable that progress has been significantly delayed by system designs that failed to focus on overall aims. It is well recognised that it is easier to enter data into clinical systems than to analyse them in order to answer questions about quality, care patterns, longitudinal trends, drug interactions, and patient safety. Although clinical information systems have slowly evolved to provide more user friendly interfaces, they still struggle with two important areas: data coding and pattern analysis. The next logical step—to mine datasets and present meaningful data patterns using visualisation techniques—has hardly been tackled. Nevertheless, researchers are slowly negotiating the rocky path from clinical data to information to knowledge.

An important inherent ability of clinical information systems is to signal possible linkages between events: to alert health professionals to be vigilant or to avoid risk. It has been postulated that it might be possible to analyse the harm that arises as a result of prescribing drugs. If patterns could be recognised, then it might be possible either to avoid the possibility or to build in safety nets to reduce potential problems. Patient datasets could be searched for possible risky combinations (co-morbidities and drug combinations) and actions taken to reduce the chance of harmful outcomes. Another realm of anticipatory care becomes available. The first step is to predict some possible “indicator” associations between prescribing and “harmful events”.

In this issue of QSHC Morris and colleagues have adapted work from the US by MacKinnon and Hepler1 who found that it was feasible to examine associations between patient morbidity and prescribed drugs by examining electronic clinical records.2 The work in North America was done on a hospital database in a managed care organisation, so it is likely that this UK article which used primary care records has a better dataset. It is also likely that the implementation of the quality indicator based contract in general practice3 will enhance the accuracy of coding in consultations over the next few years, increasing the chances that this type of work will be easier to conduct.

Three key areas of risk were found—namely, patients with heart failure or hypertension who use non-steroidal anti-inflammatory drugs, patients who are not monitored when using angiotensin converting enzyme inhibitors, and patients who use regular hypnotic-anxiolytic drugs. The authors avoid claiming that the patterns are evidence of causality—that, for instance, the use of hypnotic-anxiolytic drugs leads directly to hip fractures. Nevertheless, the development of possible temporal indicators and emergent patterns are important findings. Problems identified in the US are found in the UK and a “pareto” principle is demonstrated—a small number of possible interactions are associated with a large percentage (60%) of potentially preventable morbidity.

Drug related morbidity attributable to predicted associations was found in 1% of adult patients in the time frame studied (27 months). It is probable that under-recording in electronic records means that this is an underestimate of incident drug related morbidity. As “during consultation” coding improves, the epidemiological drug morbidity profiles achieved at practice level may well improve and potential benefits be more easily identified. Notice, for instance, the laborious data extraction, data cleaning, and repeated analysis that had to be undertaken to arrive at the results presented by Morris et al.

If this work is replicated, even if only the most common associations are correct, then knowing that these three patient groups are at significantly increased risk should lead to preventative strategies. It is therefore likely that the avoidance of drug related harm by planned review will become a future quality indicator. The authors want to use the results to generate discussion about these possible developments. They do not think that existing databases are accurate enough for comparisons between practices and are wary of such potential developments. It is possible, however, to speculate that this type of analysis might occur as the patient record becomes electronic and shared more widely. Could a detailed interrogation of associations between events and antecedent prescribing have medicolegal implications? The validity of the “association indicators” would then become essential.

Extracting and analysing electronic clinical data in order to identify temporal associations between drug prescribing and patient morbidity is an exciting step forward. At the moment it requires database analysts to sift through the codes. But it will get easier. Used in a preventative strategic way to improve patient care, the technique could be a valuable intervention for quality improvement. If, on the other hand, a more inquisitorial use comes about, it might be more difficult to harness the benefits for patients in the long term.

The use of electronic clinical data to identify temporal associations between drug prescribing and patient morbidity


Linked Articles