Design
This is a cross sectional study of secondary data originally collected for program evaluation purpose by IntraHealth International (a not for profit global public health organization) and the Namibian Ministry of Health and Social Services (MoHSS).
Setting and Sample
This study utilized 66 health facilities across 5 regions of high HIV burden in northern Namibia as the unit of analysis. Our sample comprised 7 hospitals and 59 health centers and clinics.
Overall, each facility type provides the same level of outpatient HIV services. However, some of the sickest patients are referred to hospitals which are in more populated areas if they are not responding to treatment or if they need in-patient hospitalization. Health centers and clinics, the smaller facility type, are located in community centers and often, but not always, in less populated rural areas.
Variables
Independent Variable. Nurse Staffing, the independent variable of interest, was measured using the WISN. The WISN is a ratio of the current number of nurses available in a facility relative to the number of nurses needed. The needed number of health workers is determined from the FTEs need to meet patient care demands based on observations, interviews, service statistics and chart reviews (1). The WISN ratio in this study accounted for the workload of nurses in all HIV services at the facility and outpatient visits over a one-year period. Both types of Namibian nurses, enrolled nurses and registered nurses, who initiate ART therapy were included in our WISN ratio. The WISN ratio was trichotomized into: insufficient staffing (<0.75), sufficient staffing (0.75-2.0), and overly sufficient staffing (>2.0) based on the WISN manual and expert consultation (1).
Dependent Variables. The main dependent variables in this study are US President’s Emergency Plan for AIDS Relief (PEPFAR)’s Monitoring, Evaluation and Reporting Indicator guidance variables of VLD and VLS. VLD is defined as the number of ART patients with viral load results documented in the medical record in the past 12 months divided by the number of patients on ART for at least twelve months (15). VLS is defined as the number of ART patients with suppressed viral load results (<1000 copies/ml), as documented in the medical record or laboratory information system (LIS) in the past 12 months, divided by the number of ART patients with viral load results documented in the medical record or LIS in the past 12 months (15). Patients who died or who had been lost to follow-up were excluded from the denominator.
Covariates. Several other variables were used as covariates in this analysis. The covariates used in this study were skill mix, facility type, HIV burden, patients visit at a facility each year, gender at a facility level, age at a facility level, level of poverty, average household consumption and literacy rate.
Skill Mix was measured as a percentage from the number of RNs out of the total number of ENs and RNs combined. This was included as a covariate because skill mix has been an important variable in previous nurse staffing research, with the ratio of RNs in staffing linked to several positive patient outcomes (16).
Type of facility was also included as a covariate. Type of facility was assessed as either hospital or health center/clinic. A dummy variable was created for each of these two types of facilities, with a reference group for hospitals. Type of facility has also been included in previous nurse staffing research on HIV services (17).
The next covariate is the HIV Burden defined as the HIV prevalence in the region of the facility. Previous nurse staffing outpatient research highlighted the importance of controlling for important geographical characteristics in the region of a facility (18, 19).
The number of patient visits was included as a covariate because it reflects the amount or volume of work required in a facility over a six-month period as it reflects the busyness of each facility in comparison to other facilities. This covariate did not account for patient uniqueness, meaning that a patient who came to the facility more than once were counted in this number more than once.
Analysis
This study used Poisson regression to explore the relationships between nurse staffing and VLD and VLS at the facility level. Prior to accessing any data, permission was obtained through the Institutional Review Board (IRB) at the University of North Carolina at Chapel Hill (UNC), as well as the Research Division at Ministry of Health in Namibai. The data did not have identifiable health worker or patient information on it.
Power was computed using the Poisson regression procedure of PASS 11 (22), which was based on the following assumptions: predictors are standardized, all predictors other than staffing are at their observed means, a 5% significance level, and an estimated proportion of 0.50 VLD and VLS at the observed mean staffing. A sample size of 73 provided 80% power to identify an increase of about 58% (or an increase of 0.29 units to a viral suppression proportion of 0.79), with an increase of one standard deviation in nurse staffing, in VLD and VLS. Thus, this secondary analysis of existing data was powered by identifying only relatively large effect sizes.
Modeling was conducted first by completing single predictor models; then, if WISN was significant in the single predictor models, by completing two predictor (i.e., WISN and a single covariate) models; and, last, if WISN remained significant in the two predictor models, by completing multiple predictor models (23).