Developing a User-Centered Digital Clinical Decision Support App for Evidence-Based Medication Recommendations for Type 2 Diabetes Mellitus: Prototype User Testing and Validation Study

Abstract

Background: Closing the gap between care recommended by evidence-based guidelines and care delivered in practice is an ongoing challenge across systems and delivery models. Clinical decision support systems (CDSSs) are widely deployed to augment clinicians in their complex decision-making processes. Despite published success stories, the poor usability of many CDSSs has contributed to fragmented workflows and alert fatigue.

Objective: This study aimed to validate the application of a user-centered design (UCD) process in the development of a standards-based medication recommender for type 2 diabetes mellitus in a simulated setting. The prototype app was evaluated for effectiveness, efficiency, and user satisfaction.

Methods: We conducted interviews with 8 clinical leaders with 8 rounds of iterative user testing with 2-8 prescribers in each round to inform app development. With the resulting prototype app, we conducted a validation study with 43 participants. The participants were assigned to one of two groups and completed a 2-hour remote user testing session. Both groups reviewed mock patient facts and ordered diabetes medications for the patients. The Traditional group used a mock electronic health record (EHR) for the review in Period 1 and used the prototype app in Period 2, while the Tool group used the prototype app during both time periods. The perceived cognitive load associated with task performance during each period was assessed with the National Aeronautics and Space Administration Task Load Index. Participants also completed the System Usability Scale (SUS) questionnaire and Kano Survey.

Results: Average SUS scores from the questionnaire, taken at the end of 5 of the 8 user testing sessions, ranged from 68-86. The results of the validation study are as follows: percent adherence to evidence-based guidelines was greater with the use of the prototype app than with the EHR across time periods with the Traditional group (prototype app mean 96.2 vs EHR mean 72.0, P<.001) and between groups during Period 1 (Tool group mean 92.6 vs Traditional group mean 72.0, P<.001). Task completion times did not differ between groups (P=.23), but the Tool group completed medication ordering more quickly in Period 2 (Period 1 mean 130.7 seconds vs Period 2 mean 107.7 seconds, P<.001). Based on an adjusted α level owing to violation of the assumption of homogeneity of variance (Ps>.03), there was no effect on screens viewed and on perceived cognitive load (all Ps>.14).

Conclusions: Through deployment of the UCD process, a point-of-care medication recommender app holds promise of improving adherence to evidence-based guidelines; in this case, those from the American Diabetes Association. Task-time performance suggests that with practice the T2DM app may support a more efficient ordering process for providers, and SUS scores indicate provider satisfaction with the app.

Keywords: clinical decision support; decision support; design; diabetes; electronic health record; evidence-based guidelines; type 2 diabetes mellitus; user testing; user-centered design; validation; workflows.

Figure 1

A screenshot of the prototype T2DM app. A1C: glycated hemoglobin, ASCVD: atherosclerotic cardiovascular disease, CHF: congestive heart failure, CKD: chronic kidney disease, eGFR: estimated glomerular filtration rate.

Figure 2

A screenshot of the mock electronic health record. BP: blood pressure.

Figure 3

Schematic of workflow to complete an ordering task in an EHR that includes the integrated prototype T2DM app. EHR: electronic health record, T2DM: type 2 diabetes mellitus.

Figure 4

Mock medication ordering screen in the electronic health record. DPP-4i: dipeptidyl peptidase 4 inhibitor, GLP-1RA: glucagon-like peptide-1 receptor agonist, SGLT2i: sodium/glucose cotransporter-2 inhibitor, SU: sulfonylurea, TZD: thiazolidinedione.

Figure 5

Schematic of the study design. EHR: electronic health record, TLX: Task Load Index.

Figure 6

Kano Model Survey results. A1C: glycated hemoglobin, eGFR: estimated glomerular filtration rate, EMR: electronic medical record, UACR: urine albumin-creatinine ratio.

Figure 7

Adherence to American Diabetes Association evidence-based guidelines.

Figure 8

Task times during prescribing.