The effects of clinical decision support system for prescribing on patient outcomes and physician practice performance: A systematic review CURRENT STATUS:

Background The clinical decision support systems(CDSSs) for prescription medications is one of the technologies aimed at improving physician practice behavior and patient outcomes by reducing drug prescription errors. This study was conducted to investigate the effect of various CDSSs on physician practice behavior and patient outcomes. Methods This systematic review was conducted by searching in PubMed, EMBASE, Web of Science, Scopus and Cochrane Library from January 2005 to November 2019. Two researchers independently evaluated the studies. Any discrepancies over the eligibility of the studies between the two researchers were then resolved by consulting a third researcher. Finally, data were extracted from the articles; however, we could not able to conduct meta-analysis due to the heterogeneity of the studies and the narrative form of the findings. Results Based on the inclusion criteria, 46 studies were considered eligible for the analysis in this review. The CDSS for prescription medications had been used for various diseases, namely cardiovascular diseases, hypertension, diabetes, gastrointestinal and respiratory diseases, AIDS, appendicitis, kidney disease, malaria, high blood potassium, mental disease. Meanwhile, other cases such as the concurrent prescription of multiple drugs for the patient and its effects on the above-mentioned outcomes were evaluated. The analysis shows that in some cases the application of CDSS provides positive effects on patient outcomes and physician practice behaviors. No significant difference was observed in comparison between some other cases and conventional methods. We think that this could be due to the disease type, the quantity and type of CDSS requirements that influence the comparison. Conclusions Our findings suggest that the positive effects of the CDSS are due to factors such as user-friendliness, compliance with clinical guidelines, patient and doctor cooperation, integration of electronic health records, CDSS and pharmaceutical systems, consideration of the views of physicians in assessing the importance of CDSS alerts and their real-time alerts in the prescription.


Search strategy
The initial search was conducted in PubMed in this systematic review to identify the keywords. We used Medical Subject Headings (MeSH) and other words/phrases used in similar articles as the basis of search strategy. Then, we carried out the search in PubMed, EMBASE, Web of science, Scopus and Cochrane Library. The search was conducted in 2018 using the approach designed for each database without language restrictions. Alerts have been used to access published articles after the search date. All of databases alerts were reviewed to July 2019. Also, reference tracking and citation search were used to augment the retrieval of eligible studies. An example of the full search strategy for PubMed is given below: (("Clinical decision support system" OR "Clinical Decision Support" OR "Information System" OR "hospital information system" OR "Computerized Medical Record system" OR "computer-assisted decision making" OR "reminder system" OR "computer-assisted diagnosis" OR "Clinical Informatics")) AND ("Medical Mistake" OR "medical error" OR "therapeutic error" OR "diagnostic error" OR "drug interaction" OR "drug dose-response relationship" OR "drug administration schedule").

Inclusion/exclusion criteria
This research included randomized CDSS-based clinical trial articles such as alert-based, recommendation, instructional and reminder-based systems to evaluate their effects on the behavior of patients and providers. In selecting an article for this study, we provided a list of questions whose answers are the main inclusion criteria. Here are the questions: Does the study focus on evaluating the decision support system?
Has the study been a randomized clinical trial in which patient care has been compared with and without a decision support system?
Have these studies been used by physicians, specialists, and residents in the decision support system? Does the decision support system evaluate patient-specific information in the form of management or probability options or have recommendations for physicians?
Has clinical efficacy been described as a measure of the care process, or the outcome of patients with any improvement in the study?
Screening and data extraction 5 The papers were screened in three different stages based on title, abstract and full text. Evaluation was conducted by two authors of this study (S.T) and (F.V). All phases of selection and screening of the articles were independently conducted in order to avoid bias. Any discrepancies between researchers were resolved by consulting an experts in this field (F.S).

Quality assessment
quality of studies was assessed by the Jadad scale, the Oxford research methods scoring system for bias in clinical trials [9,10]. We finally reached the conclusion that the results of the quality assessment were acceptable.

Data synthesis and presentation
The data were extracted from eligible articles using a structured data extraction process. Metaanalysis was not possible due to the heterogeneity of the studies; therefore, the results of studies were presented in a descriptive-narrative form. We used PRISMA checklist as a reporting guideline in our study. PRISMA checklist is a known standard checklist for systematic review reports [11].

Results
The evaluation of the reviewed studies provided us with enlightening information with respect to the aims of the research, types of electronic prescribing systems, illness types and patients. The findings also showed that in several diseases such as cardiovascular disease, high blood pressure and diabetes, or cases such as simultaneous prescription of drugs, the effect of CDSS was evaluated.
Findings from the analyzed studies are presented in the following tables in which * stands for p values and indicates a statistically significant difference.
Also, the results of the quest are shown in Figure 1.
The number of studies based on multiple evaluation results and types of studies are shown respectively in Figures 2 and 3.

The effect of CDSS on cardiovascular diseases
For patients admitted to hospital, the level of venous thromboembolism prophylaxis and the proportion of prescribed prophylaxis increased from six to 24 hours after admission for clinicians allocated to venous thromboembolism reminder CDSS [12]. In another study, differences among 6 physicians over the thromboprophylaxis treatment effect decreased with the help of CDSS providing treatment recommendation (p-value=0.02) [13]. In other studies, alert based CDSSs have been effective in physician behavior and progressive treatment improvement in anti-inflammatory drugs and lipid-lowering drugs, which has also been statistically significant [14][15][16]. As stated in another study, by following medical recommendations, doctors in the intervention group were able to improve the prescribing level of secondary preventive medication with the help of a regular CDSS [17].
In the other trials, the short message of the program in patient outcomes had a positive effect on patient adherence to medication, diet, and cardiovascular disease (p-value<0.01) [18,19]. Table 1 shows the result briefly.

The effect of CDSS on hypertension
For one study, the electronic monitoring and recall program had no effect on blood pressure reduction and the admission of patients [21]. However, in the other study, the patient outcome improved after the implementation of the CDSS [22]. Table 2 shows the result briefly.

The effect of CDSS on diabetes
In some studies, the Real Time Medication Monitoring (RTMM) system, equipped with a short message reminder, improved precision of patients' compliance and taking forgotten dosages [23][24][25][26]. In another study, HbA1c and group differences were greater in the intervention group using recommendation CDSS than in the control group [27]. The use of statins (p-value=0.03) and the problem areas in diabetes (PAID) (p-value=0.01) improved in another study for intervention group using CDSS [24]. Table 3 shows the result briefly.

The effect of CDSS on digestive diseases
In all studies, the CDSS had an effect on prescribing non-steroidal anti-inflammatory drugs, proton pump inhibitors, and increasing the standard use of oral rehydration solution without any difference in other results [28][29][30]. The alert based CDSS was able to improve the quality of patient care to some degree in the other study [29]. Table 4 shows the result briefly.

The effect of CDSS on pulmonary diseases
In some trials, the use of CDSS integrated with electronic health record, learning or prediction rules resulted in a decrease in the prescribing of antibiotics and macrolides, thereby minimizing inappropriate use of antibiotics (p-value=0.0005), decreasing resistance to antibiotics (p-value=0.04), and enhancing primary care [31][32][33][34][35]. The patients had adhered to the reminder message of using their medication in another study; however, the messages did not affect therapy success [36]. Table 5 shows the result briefly.

The effect of CDSS on AIDS
In the study, it was shown that the reminder systems for short text messages had a positive effect on the treatment process for antivirus. The length of the messages also required more attention from the physicians, but had no significant effect on patient compliance rates (p-value=0.12) [37]. Table 6 shows the result briefly.

The Effect of CDSS on appendicitis
The study showed that the system's systematically developed order set using clinical guidelines increased system usability (p value=0.05) and reduced system-related problems related to unfamiliarity with the system (p-value=0.05). This method resulted in the Computerized Provider Order Entry (CPOE) improving efficiency, quality and safety [38]. Table 7 shows the result briefly.

The effect of CDSS on kidney diseases
One study showed the positive effect of multipurpose intervention on creatinine value estimation and dose adjustment to reduce the insufficient dosage of primary care drugs [39]. In the other study, the appropriate prescription rate for kidney problems was rather low, contrary to the results of the study by the former. Furthermore, the effectiveness of the CDSS equipped with physician guidelines has been increased, especially for new versions [40]. Table 8 shows the result briefly.

The effect of CDSS on taking multiple medications
In one study, 194 hard-alerted CDSSs resulted in delayed drug treatment for four patients requiring immediate treatment, suggesting that adverse events of these systems need to be evaluated and monitored [41]. In another study, the CDSS improved the primary dose of medication, time intervals for drug use, and drug concentration, which is to be injected intravenously compared to standard doses [42]. Also In another study, the average number of readmission days for each patient and the 8 combination of re-hospitalization and emergency ward visits in the 30 days after hospital discharge was not different in the intervention group using Recommendation CDSS with control groups [43]. In some trials, there was no discrepancy between the outcomes of the dosage rate and the Modified Medication Appropriateness Index (MMAI), improper medication prescribing (p-value=0.48), the Medication Regimen Complexity Index and the mean pain outcome difference after 6 months (p-value=0.13) and 9 months (p-value=0.78) between intervention group using alert or reminder CDSS with control group [44,45]. Table 9 shows the result briefly.

The effect of CDSS on Malaria
The use of text messages in the study did not affect the behavior of patients in completing the course of medication for the full duration of treatment. However, if the side effects were low (p-value=0.02), it had effects in continuing to use the medication. In addition, text messages had an effect on the physicians' knowledge in using medication with fatty foods (p-value<0.0001) [46]. Table 10 shows the result briefly.

The effect of CDSS on increasing blood potassium
In one study, in following alerts and patient compliance rate, there is no statistical difference between control and intervention groups. The doctors' compliance rate improved at the medium potassium level from 3 to 3.9 (mili-equivalents/liter) (p-value<0.01) [47]. Due to the rapid response of the physicians to program reminders and alerts for high potassium levels in the intervention group, the positive effect of the system on physician behavior was evident in another study (p-value=0.01) and a high level of potassium (p-value=0.05). Thus, patient safety could be increased [48]. In another study, the time lapse in hyperkalemia monitoring (p-value=0.20) and the incidence rate of hyperkalemia (p-value=0.22) did not differ significantly even with use of three different kind of reminder and alert based CDSSs [49]. Table 11 shows the result briefly.

The effect of CDSS on medication prescription for patients
Based on the results of some studies, the regular or alert based CDSSs resulted in better drug prescriptions for the proton pump inhibitor and a reduction in abbreviation prescriptions [4,50]. In the other studies, the overall utilization of system functionalities, system utilization between two time 9 laps (p-value<0.0001), number of users (p-value<0.0001) and physician compliance regarding drug recommendations given by the CDSS improved drug prescriptions, which eventually resulted in reduced side effects (p-value=0.02) and harm to patients due to the lower number of errors regarding the alert based CDSS [5,51]. There was no difference in drug prescription between physicians in one study (p-value=0.14); However, the percentage of skill questions answered for the intervention group equipped with training CDSS (p-value=0.01) improved [52]. In another study, alert based CDSSs have been effective in identifying evidence-based pharmacotherapies (EBP) and the compliance with treatment by health care managers and have had no effect on the outcome of patients [53]. Table 12 shows the result briefly.

The effect of CDSS on mental disorders
DSS alerts resulted in reduced risk of injury and reduced dose of antipsychotics and anticoagulants (p-value=0.03) from the start of the study up to a year. Therefore, the CDSS reduced the risk of injury (p-value=0.02) [55]. Table 13 shows the result briefly.

Discussion
The aim of this systematic review is to establish the effect of CDSS on patient outcomes and physicians' performance. Figures 4 and 5 show the number of studies associated with each country and type of CDSS. The effect of CDSS was measured using different methods in included studies. In most cases, the effect of these programs on physicians' performance and patient outcomes was positive and in others was ineffective.
The results show that the use of CDSSs in cardiovascular patients has positive effects on physicians' performance through increasing the prescription of anticoagulants, anti-inflammatory drugs, antithrombotic drugs, lipid-lowering drugs, blood pressure drugs, cardiovascular drugs recommended for the reduction of cardiovascular diseases in patients with diabetes, observing clinical guidelines and improving the quality of patients [12,14,15,20] 56]. The system's userfriendliness and low running CDSS cost resulted in system efficiency in the care delivery process [12,14,15,20].
However, the study results showed that using CDSSs for cardiac patients did not affect the physicians' performance in a number of outcomes such as physician conduct in prescribing drugs, the warfarin treatment system, and minimizing dissatisfaction with guidelines for antithrombotic diagnosis and job satisfaction [13,[15][16][17]. The results of this study are consistent with studies by Byrnes and Lazaro who discussed that clinical factors and treatment issues were the reasons for physicians' disagreement with system recommendations [57,58]. The main reason that no change was found in medical guidance disputes was the difficulty of clinical conditions which could increase the risk of patient injury and hinder the decision-making process [13,[15][16][17].
The results of this study indicate that the use of CDSSs in cardiovascular patients has a positive effect on a number of outcomes such as adherence to drug use by patients and following a nutrition-based diet in the Mediterranean [18,19]. Similarly, according to clinical guidelines and reminders, the study by Schedlbauer et al. evaluated the effect of CDSS on cardiovascular patient outcome as positive [59]. The reason for low Mediterranean diet adherence was the delivery of a short message outlining the advantages of the Mediterranean diet, which resulted in an improved conformity level [18,19].
The study also showed that the use of CDSS in cardiovascular patients did not affect patient outcomes such as readmission rate, mortality or smoking cessation [18,19]  The study results indicate that the use of CDSS for appendicitis has a positive effect on the physicians' performance in certain outcomes such as performance, quality and safety with the aid of the physician's computerized order entry [38]. The results of this study are in line with Holden's report, which explores how doctors who use the order entry system get more easily up-to-date information and boost the system's ability to use it [71]. Although prescriptions are not strong in terms of content, errors are reduced as they cause doctors to think about the cases and the data are not selected automatically [38].
The results also show that use of CDSS in kidney patients has a positive effect on the behavior of the physician in some outcomes such as reduced dosage of inadequately prescribed drug and improved rate of adequate prescription [39,40].  [72,73]. Timeliness of the CDSS alerts was among reasons for the success of CDSS in prescribing adequate drugs and correct dosage [39,40]. Results of the study showed that use of CDSS in prescribing drugs for patients has a positive effect on physicians' performance in certain outcomes such as drug prescription for proton pump inhibitors, CDSS productivity and usability, reducing side effects of drugs, and improving the learning rate and skills of physicians [4,5,[50][51][52]. The results of this study are consistent with the results of Curtis and Shah et al. which indicated that relevant CDSS, while providing users with performance-related information, reduces patients' harms and errors, and increases the physician's enhanced knowledge and skills [76,77]. One of the main reasons for the proton pump's enhanced medication performance was the monitoring of the physician's prescribed drug dose, as well as equipping pharmacies with CDSS with hard alerts which reduced costs and improved usability [4,5,[50][51][52].
Results show that use of CDSS in prescribing a number of drugs has a positive effect on the physicians' performance in some outcomes such as number of emergency ward visits, number of rehospitalizations, determination and supervision of the amount of drugs including the initial dose [42,43]. The results are consistent with Vincent and Cordero's research, which demonstrates that combining the computerized order entry process with an alert system saves time in prescribing and optimizing the dosage of drugs [78, 79]. The reason for CDSS ' positive effect on the number of rehospitalizations, emergency ward visits, and reduced morbidity rate was that CDSS had knowledge base in pharmacogenetics and was equipped with drug interaction alerts [42,43].
Analysis of the results of the reviewed studies shows that use of CDSS in prescribing a number of drugs has no effect on the physicians' performance in outcomes such as drug prescription rates with drug suitability index and average functional status outcome and drug complexity index [41,44,45].
Our study results are consistent with Olsson, suggesting that CDSS for elderly people who use multiple types of medicines has no effect on important outcomes when prescribing medicines [80].
The explanation for the incidence of unexpected findings in this research was the lack of information for patients with serious infections requiring immediate care and the lack of relevance of the checklist 14 given for the drug problems of the patient [41,44,45].

Conclusion
This systematic review study was conducted with the aim of identifying the effect of CDSS on patient outcomes and physicians' performance. The results show that use of CDSS in some diseases has positive effects on the outcomes of patients and the physicians' performance as well, while it has no significant effect on others. In addition, the type of outcomes and the effects of CDSS on the disease are different. Using this technique in some cases yields positive results in patients and doctors, while in others it demonstrates no significant difference compared to those of conventional methods. The positive effect of CDSS seems to be due to factors such as user-friendliness of the system, the number of patients requiring treatment, the rate of observance of clinical guidelines, conformity of clinical guidelines and data registry, the rate of patients' accurate adherence to messages of the system, useful short messages, the existence of algorithms with dynamic functioning based on patient data, existence of patient medical records, the relationship between electronic health records with CDSS and timely alerts of the system in the prescribing process. In addition, the positive effect of CDSS depends on a number of other factors such as having an instruction section, not being confronted with mandatory recommendations, patient and physician cooperation with the help of CDSS guidelines, not lagging between alerts where the alert is of low importance and the identification of important alerts, equipping pharmacies with CDSS and system applicability, considering the opinions of doctors when assessing the value of alerts and notifications for drug interaction.

Limitation
The selected papers in this study were about different diseases; therefore, we had to have different process design, results and evaluation criteria which made it impossible for us to conduct a metaanalysis. The results are therefore presented in a narrative form.

Ethics approval and consent to participate
Before collecting the data for the present study, authors obtained the Code of Ethics from Ethics Committee for University Research, Iran University of Medical Sciences.
All stages of research were conducted by two independent researchers.
During the research project, all papers related to the accuracy of the method were identified and analyzed.

Consent to publish
There is not any individual person's data in any form in this research.

Availability of data and materials
All data generated or analysed during this study are included in this published article [and its supplementary information files].

Competing interests
The authors declare that there is no financial and non-financial competing interests associated with this research.

Funding
There was no funding resource for this study.              Figure 1 Algorithm of screening articles based on Preferred Reporting Items for Systematic Reviews The number of studies based on several evaluating outcomes 41 Figure 3 Type of included studies The number of studies associated with each country 42 Figure 5 The number of studies associated with each CDSS type