The positive findings in 15 of 20 studies167,178,228–232,234,236,237,239,240,242,246,247 were encouraging. Several further studies provided results that did not provide insight into how successful the CDSS was.237,238,251 However, proof of positive benefit on actual patient outcomes that can be attributed to the CDSS was limited.232,234 Most of the studies investigated improved prescribing or investigative tests, and about half of studies did not examine CDSS versus control.167,178,229,230,232,238,239,242,251
Several studies focussed on identifying DDIs, or where additional data was available, drug appropriateness and dosage in the patient population (notably: excess use of psychotropics
Clinical decision support systems
and anticholinergics in the elderly).225,228–232 Studies investigating specific disease management looked at guidelines for test frequency, lifestyle management and appropriate drug and dosage use.233–236,238–240,242,246–251
One study noted an excess of clinically insignificant alerts produced by CDSS231 which was also suggested by three others.248–250 The authors of the Tamblyn study231 suggested patient context was lacking, with similar findings concerning the application of simple guidelines in other studies.248–250 Two thirds of GPs in the Sequist study247 reported not noticing the reminders, which may have been in some part due to alert fatigue. Whilst details of the types of approaches, or rules used, for the various CDSS are unclear, many studies appeared to use simple conditional rules.228,230,231,233,234,236,240,247–251
One study using Janus provided results of CDSS alerts of clinically significant incidences, but did not say how often insignificant alerts occurred.225 The Swedish Janus application was previously identified with excessive alert production by Mannheimer et al.253, yet no mention of this effect was provided in the two reviewed studies.225,232 Mannheimer noted clinical monitoring context was not incorporated and would have been likely to reduce the high number of clinically unimportant DDI alerts issued, again highlighting the issue of alert fatigue (Janus employed a severity coded DDI database253).
Although the Janus application classified alerts by severity, and the Tamblyn study231 allowed GPs to modify alert sensitivity, the issue of displaying reminders and alerts in the context of patient health factors remained.
Eight studies involved the use of CDSS for the general medication review process.167,178,225,228– 232 Most of these studies targeted patients who would have been likely to benefit from medication review, being the elderly or hospitalised patients.225,228–230,232 Of the studies225,228–232 measuring the effect of the CDSS intervention, five found positive benefit.228–232
Many studies utilised CDSS for specific disease management.233–240,242,246–250 Many of the diseases studied (CHF, hypertension, CAD, angina, diabetes, asthma, COPD, lipid management), were more common conditions which typically involved the use of complex management guidelines. Many of these conditions have been associated with evidence- practice gaps, CHF254, CAD28,29, asthma40,255, lipid management29, diabetes.29,256 The
Clinical decision support systems
complexity of management suggested CDSS could have a positive impact in the management of patients with these conditions, yet of the eleven studies233–236,240,242,246–250 which investigated guideline adherence or patient outcome, only six showed at least some positive benefit.234,236,240,242,246,247
Half of the studies228,231,233–235,246–250 used a randomised controlled trial (RCT) design which provided a more rigorous basis for comparison of the effect of CDSS intervention, with or without any associated interventions, of which six produced some positive results.228,231,234,240,246,247
The three studies associated with the Indiana University research group used the Regenstreif Medical Record System (www.regenstreif.org) resulted in virtually no improvements in guideline adherence or patient outcomes.248–250 Tierney suggested the reminders were invasive and time consuming. Tierney also stated GPs could easily skip past the reminder by pressing a key rather than read and provide a response.248 The subsequent studies by Tierney and Subramania may have also had this limitation.249,250
Six167,178,232,239,242,246 of eight167,178,225,232,235,239,242,246 studies using knowledge based systems showed positive results. Two studies of knowledge based systems did not assess change in terms of patient or healthcare practitioner outcomes.237,238 Six knowledge based system studies were either pre-deployment or prototypes.167,178,237–239,242 The advantage of knowledge based CDSS is the capacity of the system to integrate a wide range of patient contextual information to provide a more considered response. This is shown by the majority of positive studies of expert systems found in this review. Unlike simple conditional rules which were apparently used in many of the reviewed studies, expert systems may be expected to be able to provide more suitable, patient-contextual alerts and reminders, minimising the excessive alerting effect mentioned or suggested in several of the studies.231,248–250
An interesting observation was of three studies that utilised knowledge based systems. These were found to be more thorough (or perhaps more consistent) in detecting DRPs than their human counterparts.167,178,239
Lack of uptake by health professionals of the trialled systems was noted in several studies.235,246–248 This lack of uptake also could have affected other studies, though no clear
Clinical decision support systems
mention of this was made. Several reasons for lack of uptake were discussed in some of the studies: CDSS advice was too simplistic, advice was not tailored for the needs of individual patients, CDSS advice was passive and as a result easy to ignore and practitioner reluctance to change existing processes.
There have been other articles which have discussed the possible lack of effectiveness of CDSS in general.153,257,258 More than half of the studies examined support the effectiveness of CDSS, particularly when implemented with rules that incorporate individual patient detail. Utilising such detail has provided conclusions of greater complexity when drawn from such individualised patient context, particularly as can occur with expert systems.