Understanding HIV Care Provider Attitudes Regarding Intentions to Prescribe PrEP

Amanda D. Castel, MD, MPH; Daniel J. Feaster, PhD; Wenze Tang, MPH; Sarah Willis, MPH; Heather Jordan, MPH; Kira Villamizar, MPH; Michael Kharfen, BA; Michael A. Kolber, MD, PhD; Allan Rodriguez, MD; Lisa R. Metsch, PhD


J Acquir Immune Defic Syndr. 2015;70(5):520-528. 

In This Article


Survey Administration

Surveys were administered between March 2012 and March 2013 to HIV providers in Washington, District of Columbia and Miami-Dade County, Florida, assessing provider knowledge, attitudes, and beliefs regarding PrEP and who should receive it, as well as perceived barriers and facilitators to PrEP provision. Survey methods varied slightly between cities, but identical survey items were administered in both cities.[22] The target study population included infectious disease providers and HIV providers who had treated at least 1 HIV-positive patient in the previous year. Listings of HIV providers from physician societies, training centers, and health departments in both cities were used to identify potential participants.[22] A brief, internet-based, anonymous survey was administered to 124 providers in District of Columbia and 107 HIV providers in Miami-Dade County using Research Electronic Data Capture (REDCap, Vanderbuilt University, Nashville, TN) and Survey Monkey, respectively, or a mailed hard copy survey. In both cities, providers received periodic mail, telephone, and e-mail reminders to encourage participation. Online or written informed consent was obtained and providers completing the survey received a $20 incentive. IRB approval was obtained from the George Washington University, the District of Columbia Department of Health, the University of Miami, and Columbia University.

Survey Domains and Analytic Methods

Survey requests were sent to 231 providers; 142 providers (District of Columbia, n = 63; Miami, n = 79) responded (overall response rate 61%). Data from the 2 cities were subsequently merged and aggregated for analysis purposes.

PrEP Knowledge/Experience

There were 5 questions regarding knowledge of and experience with PrEP (familiarity with iPrEX results[1] CDC guidelines,[23] practice having written PrEP protocols in place, frequency of PrEP requests, and ever having prescribed PrEP). These questions were combined into a single, "lack of PrEP knowledge/experience" scale with higher values indicating less knowledge/experience. This scale had an implied composite reliability of 0.83.[24]

Patient Factors Associated With Intended PrEP Prescription

The survey assessed how several factors might influence providers' decisions to prescribe PrEP, on a scale from 1 ("least likely to prescribe") to 5 ("most likely to prescribe"). Variables included whether patients had: multiple sex partners; history of failing to use condoms; partners with known HIV; history of sexually transmitted diseases; history of noninjection drug use; history of injection drug use; history of not returning for medical visits; and history of medication nonadherence. Although this survey question did not specifically ask about PrEP prescribing among men who have sex with men, it was presumed that this population would be captured through the other patient populations included in the survey question. These factors were combined into a "likelihood of prescribing PrEP" scale, with higher values reflecting higher likelihood of prescribing PrEP. This scale's composite reliability was 0.94.

Provider Perceptions, and Intentions Regarding PrEP

LCA was used to classify providers based on their attitudes toward prescribing PrEP. Nine survey items were used to identify the latent categories. Each variable was coded, so that lower scores represented the "lowest likelihood to prescribe PrEP." Latent class indicators included 2 items asking providers to rate on a 1–5 scale the effectiveness of oral PrEP and vaginal microbicides, gels, and creams in preventing HIV transmission. Another 7 items measured providers' level of agreement using a Likert scale ranging from 1—strongly agree to 5—strongly disagree with the following statements: (1) it is feasible to provide PrEP in practice; (2) there is adequate time to provide PrEP in practice; (3) PrEP will promote HIV resistance; (4) PrEP will promote risky behavior; (5) I will provide PrEP to HIV discordant couples; (6) the availability of PrEP may empower women who are unable to negotiate consistent condom use with their partners; and (7) the cost of PrEP will still be a significant barrier for those who may benefit, even if PrEP is safe, efficacious, and made available.


The following covariates were examined to identify the provider characteristics of each of the latent classes: age, sex, race/ethnicity, years of practice, field of practice, number of patients seen in the clinician's practice in the previous month, number of HIV patients seen in the clinician's practice in the previous 3 months, number of HIV patients seen by the clinicians in the previous 3 months, and the "lack of PrEP knowledge/experience" and "likelihood of prescribing PrEP" scales.

Rationale and Methods for the LCA

LCA aims to identify subgroups of individuals who respond differently on a series of categorical or ordered categorical variables. We used this method to see whether there were distinct subgroups of providers with different response patterns. The LCA was conducted using Mplus version 7.[25] First, we determined the number of classes to include by comparing the fit of models including different numbers of classes. We used Bayesian information criteria,[26] Akaike information criteria (AIC),[27] the sample size adjusted Bayesian information criteria (ABIC),[28] and entropy statistic as the criteria to compare model fitting with different numbers of classes assigned. To avoid having local maxima, each model was originally estimated with 500 starting values, with the 50 runs with the highest likelihood after 20 iterations continued to full maximization. If the maximum likelihood solution was not repeated numerous times in the set of 50, this was raised to 1500 initial starts and 500 to completion, and then 3000 and 1000, thereby ensuring that a substantial proportion converged to the same maximum. Because of the missing data, the LCA was completed with multiple imputation with 30 sets of imputed data and results combined within Mplus.[25]

Survey participants' demographics and practice characteristics were described using univariate analysis. The latent class solution is described by the unconditional probability of each class and the conditional probability of endorsing a 4 or 5 on each of the 9 input items for each class. We used a classify–analyze strategy to compare the other variables by latent class membership, which is most appropriate with high entropy models. Entropy with values approaching 1 indicates clear delineation of classes.[29] For categorical variables, χ2 tests were combined across imputations using the method described by Li et al,[30] which results in an F-statistic and associated P value. The equality of our 2 scales across classes was tested using Proc MIanalyze in SAS 9.3, which results in a t test statistic. In tables, we present the observed data means and frequencies; however, all overall statistical tests reported are based on the combined, multiply imputed data. We report the significance of the individual items composing the scales based on the observed data using a χ2 statistic.