Health care providers use a variety of computerized medical information sources to reduce knowledge gaps and support patient care decisions. Few studies have evaluated the impact of medical information sources on patient outcomes. Skin problems are the reason for many visits to primary care providers and result in a high percentage of referrals to dermatologists and return visits to primary care for the same skin problem.

The objective is to evaluate the impact of primary care providers' use of a dermatology information source, VisualDx, on skin problems outcomes.

The study design is a cluster-randomized controlled trial. Participants include primary care providers as clusters and their patients with skin problems. Providers are randomized to intervention group that refers to VisualDx when seeing a patient with a skin problem, or to the control group who does not. Patients have the randomized group status of the doctor they saw for the problem.

Patients are interviewed to determine the problem status and how many follow-up visits they had for the problem at intervals after the index visit.

The research design is a cluster randomized controlled trial of a computerized dermatology information source. This is a two-level design in which primary care providers (PCPs) who see patients for skin problems are the subjects of randomization. Their patients are the subjects of the data collection and analysis. The cluster design examines how much variation in the outcomes is attributable to the providers, how much to patients within the clusters, and how much to the intervention information source itself.

Provider Level:

After completing a baseline questionnaire, providers are randomized to intervention or control groups. Intervention group providers have an orientation to the study protocol that includes how to access and use VisualDx. Control providers receive the protocol overview, but no orientation to the intervention.

Intervention providers agree to refer to VisualDx content when they have a patient with a skin complaint. Control group providers agree not use VisualDx as a reference but may use another medical information source or none.

Patient Level:

To identify eligible patients, authorized investigators reviewed information from the patient record sufficient to send a letter to them over the provider's signature to notify them about the study. The letter gives the option to refuse any contact.

At 30 days after the primary care visit, a study team investigator phones each eligible patients, explains the study, and obtains verbal consent. The investigators are blinded to the randomized group status of the patients.

The investigator proceeds with the interview questionnaire on consent. Age, sex is collected. Patients are asked whether the skin problem is resolved ("Is it all better?"). For every resolved problem, we ask when (date or days since visit) it was resolved. We also ask how many return visits the patient has had with any provider for that problem.

If not resolved, the patient is called again at 60 days and asked the same questions. If not resolved then, we call the patient again at 90 days. The last call to any patient is at 90 days from the index visit for the problem.

Data Analysis and Statistical Measures

The primary predictor variable for both aims is the use of the VisualDx information source by the primary care provider.

For the "days to resolution" outcome variable, the time unit (days) and the status of the problem after 90 days (resolved or not resolved) will be evaluated for each group with Kaplan-Meier survival analysis and the non-parametric log rank test of significance.

For the "number of follow-up visits" outcome variable, the binary predictor (intervention group) and discrete outcome variable (number of visits) will be evaluated with Wilcoxon-rank sum test for non-normal data. The first visit of a patient to a participating provider for a skin problem is considered the index visit and that provider considered the index provider regardless of the provider seen on any subsequent appointment. Analysis will adhere to the intention-to-treat principle with group assignment determined by randomization regardless of whether VisualDx was deployed for care or not. Problems unresolved at the last contact will be censored as unresolved for the Kaplan-Meier analysis.

For aim 1, the estimated sample size to detect an average difference of 8 days in time to resolution with a standard deviation of 20 and an intra-cluster correlation (ICC) of 0.025. With alpha .05 and beta .80, the study would need a minimum 26 providers enrolled with an average 10 patient participants in each provider's cluster for a total of 286 participants completing the study. We planned to enroll at least 30 providers and up to 500 patients to insure against loss to follow-up of, increased standard deviation, or larger than anticipated Intra-Cluster Correlation.

Inclusion Criteria:

• Patients seen by a participating Primary Care Provider for a chronic or acute skin problem

Exclusion Criteria:

• Skin problems due to burns or lacerations
• Cognitively impaired, mentally ill, prisoners, non-English speaking

Adverse event monitoring began on recruitment of provider participants through completion of patient interviews: August 2015 through October 2016, approximately 14 months.

We hypothesized that there were minimal risks to participants such as loss of time spent learning new information or answering questions by providers and loss of privacy by patients, but protected health information was only exposed to treating physicians, the Principal Investigator and study team. Other hypothesized minimal patient risks included the bother of phone calls and loss of time spent answering questions. Adverse events were monitored for both patient and provider participants.