Medication reviews of hospital patients aged 65 or over were undertaken by nurses briefly trained in pharmacology compared with control in a Swedish study by Bergqvist et al.225 The Janus web application226 assisted in calculating creatinine clearance and drug-drug interactions (DDIs). Outcomes investigated were re-admission within three months, inappropriate drug use (anticholinergics, long acting benzodiazepines, multiple psychotropics) and the number of DRPs detected. No significant differences were found for
Clinical decision support systems
readmission or inappropriate drug use. Utilisation of the CDSS identified 86 clinically significant DRPs in 53 patients, which would have otherwise been missed. The number of clinically insignificant alerts was not mentioned.
Bindoff et al.167 developed a prototype CDSS expert system, using MCRDR technology227, specifically for patient medication review. Patient medication review data for 126 real cases were sequentially entered into the expert system by a medication review accredited pharmacist. This resulted in 250 rules which included information of patient symptoms, demographics, medical history, medications and pathology results. The expert system was able to identify 80% of the potential problems detected by the medication review pharmacist. Less than 10% of the problems identified by the expert system were incorrect, reducing to 0% during entry of the final 15 cases. Based on the entered rules which covered 80% of potential DRPs detected by the pharmacist, the expert system identified at least one more DRP than the pharmacist per case entered. This suggested the pharmacist missed problems even though the pharmacist knew about the same problems, and had at some point entered a rule into the expert system for these problems. Interestingly, the pharmacist routinely identified problems with later cases that they had previously missed on earlier cases.
A subsequent article by Bindoff et al.178 discussed the progression of the prototype CDSS. Improvements were enacted in the way medications and medical conditions were portrayed enabling the pharmacist expert training the system to develop suitable rules with greater ease. The revised system was trained with 244 patient cases resulting in 383 rules which covered 90% of the potential DRPs detected by the pharmacist. The authors concluded that the pharmacist expert in the study, and likely experts in general, do appear to routinely miss relevant DRPs and this type of software can help to reduce this issue. A limitation of this finding was the use of only one pharmacist expert.
Another Swedish study by Bladh et al.228 investigated CDSS-assisted hospital medication reviews vs. control, assessing patient quality of life, DRPs and drug appropriateness. Reviews were assisted with use of CDSS which incorporated electronic guidelines for DDIs and drug appropriateness in the elderly. The CDSS was not well described but may have been similar to the Janus application. Patient self-rated health was improved using a post-six month analysis of the medication review intervention. Health-related quality of life was improved in
Clinical decision support systems
the intervention group mean 3.14 ± 0.87 compared with the control mean 2.77 ± 0.94, p = 0.02, using a one to five rating to the question “In your opinion, how is your state of health?” Other self-rated quality measures were not significant using EuroQol-5 dimensions (EQ-5D) or EuroQol Visual Analogue Scale (EQ VAS). A reduction was found in the prescribing of three or more psychotropic drugs per patient (percentage of patients: control: 9.4%, intention to treat analysis: 7.9%, per protocol analysis: 2.3%, p = 0.034). Whilst CDSS was involved in this trial, it may not have been the only factor leading to the positive results. The intervention group included additional interventions of medication discussions with patients and medication reports sent to patients' GPs. The positive results shown in this study may not be entirely due to the use of CDSS.
Software, titled Pharmanurse, was developed to assist nurses to identify ADRs as a component of medication review of nursing home residents.229 The software presented nurses with resident-individualised lists of potential DRPs for 418 residents from eight nursing homes. The nurses identified a mean of 3.7 DRPs per resident of which GPs agreed in 54% of DRPs implementing 214 medication changes in 88 residents. Health professionals were generally satisfied with Pharmanurse which scored 7 out of 10 for the potential to improve pharmacotherapy. As with the study by Bladh et al. the DRPs identified and changes implemented were not entirely due to the CDSS.
In a year-long trial by Monane et al.230, 2.3 million elderly patients were automatically screened according to Beers criteria to initiate a short medication review. Pharmacists were alerted to 43,007 potential problems. Subsequently, pharmacists were able to contact 19,368 GPs regarding 24,255 alerts to discuss potential DRPs and recommendations for resolution. Change of therapy was determined through GP telephone contact and through analysis of subsequent patient prescription claims. Change of therapy to more suitable medications occurred at differing rates depending on the class of medication. The largest medication groups resulted in significantly changed prescribing: 24% change for most long-acting benzodiazepines (11,344 alerts, p<0.001), 40% change for flurazepam (1,679 alerts, p < 0.001), 25% change for shorter-acting benzodiazepines exceeding the recommended daily dose (4,532 alerts, p < 0.001), 17% change for anticholinergic antidepressants (2,856 alerts, p
Clinical decision support systems
< 0.001). The strength of this study was the very large number of patients involved, finding a small proportions of DRPs which resulted in substantial therapy changes.
Tamblyn et al.231 compared the effect of an active computer-triggered CDSS alert against a passive physician activated analysis alert by comparing follow-up prescription information. CDSS was triggered during CPOE as well as when a patient chart was opened. Alert sensitivity settings were available to either group and could be adjusted in display sensitivity by physicians in either group. In both groups not all physicians actively used patient EHRs reducing the opportunity to detect changes. The active CDSS identified 6,505 problems of which 668 alerts were displayed to the physician and 81 prescribing problems were revised. The passive physician activated alerts identified 4,445 problems of which 41 were seen by physicians and 31 were revised. At the end of the follow-up period no difference in the number of prescribing problems was identified, yet there was a significant reduction in the number of therapeutic duplication problems (odds ratio (OR) = 0.43, 95% confidence interval (CI) 0.29 – 0.64, p < 0.001). It is of interest to note the active CDSS alerts were ignored in 88% of alerts. Physician reasons mostly concerned lack of benefit or clinical irrelevance. The high number of false positive alerts, even with adjustable alerting sensitivity was of interest, as it highlighted the issue of providing patient contextually relevant alerts, which was suggested by the authors as a cause of alert fatigue and for future improvement of the software.200,201
Another Swedish study by Ulfvarson et al.232 utilised the Janus software for medication reviews in 233 elderly patients. Medication review included pharmacologist opinions and CDSS advice concerning DDIs and medication appropriateness from patient data entered into CDSS software. Measurements were conducted for DRP frequency, dosage changes, number of drugs used, and frequency of inappropriate drug use before and after. Evaluation of the CDSS itself, or its impact on the medication review process, was not investigated. Initial review found a mean of 10.4 drugs and a mean of 1.5 DRPs per patient. Three or more psychotropic drugs were used by 34% of elderly patients. The pharmacologist issued a mean of 3.3 physician recommendations per patient. A follow-up at two months found a mean 9.5 drugs per patient was observed. Reductions were found for: drugs associated with kidney impairment (17%), anticholinergics (40%), long-acting benzodiazepines (17%), drug
Clinical decision support systems
duplication (30%), and use of three or more psychotropics (19%). At two months the adoption of recommendations did not affect patient health in 213 patients and improved health in 51 patients, yet worsened health in 30 patients as assessed by researchers and health care staff.