We invited healthcare providers and clinic administrators from Title X-funded family planning clinics in Department of Health and Human Services (DHHS) Regions III, IV, and VI to participate in an online survey from February through June 2018. These regions include states that comprise the Southern US including Alabama, Arkansas, Delaware, District of Columbia, Florida, Georgia, Kentucky, Louisiana, Maryland, Mississippi, New Mexico, North Carolina, Oklahoma, Pennsylvania, South Carolina, Tennessee, Texas, Virginia, and West Virginia. Quantitative surveys took place online with the assistance of the National Clinical Training Center for Family Planning (NCTCFP). The survey was emailed to the Title X clinic listserv for DHHS Regions III, IV, and VI, and listserv members received one to two email reminders per month. An advertisement for the survey was also posted on the NCTCFP website, and the survey was promoted though engagement with State Title X Grant holders and through in-person recruitment at the biannual NCTCFP national meetings for Title X providers. The survey was guided by CFIR constructs relevant to PrEP implementation. It also collected information on readiness to implement PrEP as well as individual and clinic characteristics. The survey took approximately 15-25 minutes to complete, and participants received a $30 incentive. All procedures and recruitment methods have been described previously (33). Institutional Review Boards from Emory University and University of North Carolina approved the study protocol prior to data collection (See Additional File 1 for STROBE checklist).
Providers were defined as individuals who have the potential ability to prescribe, counsel, or screen for PrEP. Clinic administrators were individuals who served in an administrative oversight capacity over the Title X activities in their clinic. Clinic addresses were geocoded to identify participants residing in the same clinic, and each participant and clinic were given a unique identification number. Procedures for geocoding clinics have been described elsewhere (22).
There were 742 individuals who agreed to complete the survey. Of those, 519 (69.9%) respondents sufficiently completed the survey to warrant inclusion in our analyses. Our analyses used data from all surveys that met a minimal criteria for completeness. In particular, we required the respondent to provide at least one response to questions related to PrEP use in their clinic. For the purposes of this pre-implementation analysis, the sample was restricted to clinics where PrEP was not currently being prescribed. Therefore, our sample included 414 participants from 227 non-PrEP prescribing Title X clinics. On average, there were 1.8 (range: 1-12) participants per clinic.
The mean age of participants was 45.92 (SD=11.22) years. Participants were predominately White (n=311, 75.12%) and most were clinic providers (n=351, 84.78%). The mean number of years serving in their current role was 8.58 (SD=7.89). Many clinics were located in metropolitan areas (n=293, 70.77%) with high prevalence of HIV (mean=459.78 cases/100,000, SD= 512.89 cases/100,000). See Table 1 for characteristics of the sample.
CFIR Constructs. The CFIR (21) provides a menu of constructs that can be used as a practical guide for systematically assessing potential barriers and facilitators in preparation for implementing PrEP. Because it is often not practical or necessary to assess all constructs in a single study, evaluations typically focus on a subset of constructs. Constructs for this analysis were selected based on their likelihood of: 1) being a potential barrier (or facilitator) to PrEP implementation and 2) having sufficient variation across clinics. For this analysis, nine different CFIR constructs were considered, based on a literature review and discussions with experts and providers in the field of HIV prevention and implementation science. The constructs covered four CFIR domains, including (1) intervention characteristics (i.e., complexity, relative advantage, and cost), individual characteristics (i.e., staff’s attitudes), inner setting factors (i.e., implementation climate, compatibility, leadership engagement, and available resources), and outer setting factors (i.e., cosmopolitanism). All CFIR-related survey items were evaluated on a 5-point Likert scale (1= strongly disagree to 5= strongly agree). Composite scores for each construct were calculated by taking the average of the contributing survey items.
Complexity was measured using one item: “providing PrEP at my clinic seems easy to do”. This measure was reverse coded, so that higher scores indicate higher complexity. Relative advantage was measured using one item: “PrEP would be more effective than interventions we are currently promoting to prevent HIV among patients at our clinic.” Cost was measured using five items such as “PrEP is too expensive” (Cronbach’s alpha= 0.62). Attitudes were measured using a composite score of 10 items, such as “I am concerned that PrEP is not effective” (Cronbach’s alpha= 0.82), where higher scores indicated more concerned or negative attitudes about PrEP. Implementation climate was measured using a composite score of five items, such as “individuals working at my clinic value new types of HIV prevention practices” (Cronbach’s alpha= 0.83). Compatibility was measured using a composite score of three items, such as “PrEP seems suitable for patients at my clinic” (Cronbach’s alpha= 0.88). Leadership engagement was measured using a composite score of three items, such as: “senior leadership/clinical management in my clinic reward clinical innovation and creativity to improve patient care” (Cronbach’s alpha=0.90). Available resources was measured using a composite score of four items, such as “we have the necessary support in terms of staffing” (Cronbach’s alpha= 0.83). Cosmopolitanism was measured using one item: “individuals in my clinic are connected with other community organizations that provide HIV prevention services to patients.”
Readiness for Implementation. The primary outcome for this study is readiness for implementation of PrEP, which was measured based on steps of the PrEP delivery cascade (i.e., Step 1: HIV risk assessment, Step 2: PrEP education, Step 3: PrEP eligibility assessment, Step 4: PrEP prescription, and Step 5: PrEP follow-up and monitoring) that the clinic could confidently implement. This measure is derived as a composite score based on the 19 or 23 (provider vs. administrator versions) survey items (Cronbach’s alpha=0.92). Survey items include “others in my clinic can counsel a patient on the potential side effects of PrEP” and “others in my clinic can help patients navigate insurance payments regarding PrEP treatment.” Responses to each of the survey items follow a Likert scale (1= strongly disagree to 5= strongly agree). Readiness for implementation was defined as the average score from the contributing survey items.
Demographic and Clinic/County Characteristics. Individual and clinic-level characteristics were also assessed in the survey. Individual-level characteristics included self-reported age, race (white or non-white), ethnicity (Latinx or not Latinx), role (administrator or provider), years in role, ability to prescribe medicine (yes or no). Clinic-level characteristics included metropolitan location (yes or no), DHHS region (III, IV, or VI), pharmacy on-site (yes or no), staff to provide insurance support on-site (yes or no), presence of primary care services (yes or no). Clinic urban-rural status was defined using the 2013 NCHS urban-rural classification scheme for counties where Metropolitan (urban) includes large central, fringe metro, medium metro, and small metro; and Nonmetropolitan (rural) includes micropolitan and noncore counties (34). Additionally, clinic county characteristics included: HIV prevalence (per 100,000), percent of population under 200% of the federal poverty level, percent uninsured, percent White, percent Hispanic, percent of women aged 15 to 44 (childbearing age) based on AIDSVu (HIV prevalence rate only) (35) and U.S. Census data (36).
Missing data. The number of missing observations per variable ranged from 0 to 35. We used a non-parametric missing value imputation for mixed-type data (i.e. continuous and categorical) to impute missing values for all variables described above (including CFIR constructs, readiness for implementation, and demographics and clinic/county characteristics) (see “missForest” package for R (37)).
Descriptive statistics were performed on all survey items (means (SD) or counts (%)). Pearson correlations were performed between all CFIR constructs to test for multicollinearity (see Additional File 2 for correlation matrix). Most CFIR measures were significantly correlated with each other, however, there were no strong correlations (all Pearson correlation coefficients are < |0.50|), so multicollinearity was not a major concern.
Using the nine CFIR constructs (complexity, relative advantage, cost, attitudes, implementation climate, compatibility, leadership engagement, available resources, and cosmopolitanism), LPA was performed in R using the “TidyLPA” package (38), to determine if Title X clinic providers and administrators coalesced into distinct sub-groups. We investigated solutions with one to ten groups. A one group solution would assume that all providers and administrators had similar perceptions across all CFIR measures, which was unlikely; and we assumed that solutions with more than 10 groups would have produced groups that were too small for generalization. To select the number of groups that fit the data best, we used four statistical metrics as well as interpretability of the groups. We considered solutions with high entropy levels (values >0.80 indicate a high level of separation between the groups (39)), significant p-values for the bootstrap likelihood ratio test (BLRT) (which tests if the model performs significantly better than the K−1 group solution), and comparatively low AIC and BIC values. Simulation studies have shown that BIC and BLRT perform most reliably in latent profile analyses (40).
Based on the probabilities of class membership, each participant was assigned a group. To understand the composition of each group, bivariate analyses were performed between group membership and individual/clinic-level characteristics (chi-squared was used for categorical variables and one-way ANVOA for continuous variables). We then conducted a random-intercept, multilevel model to test whether group membership predicted readiness for implementation of PrEP, controlling for individual and clinic-level factors. This analysis modeled the nesting of participants within clinics. We fit the model to our data with SAS PROC GLMMIX using maximum likelihood estimation.