The Impact of Quality Assurance Initiatives and Workplace Policies and Procedures on HIV/AIDS-Related Stigma Experienced by Patients and Nurses in Regions with High Prevalence of HIV/AIDS

The original study, a prospective quasi-experimental study evaluating the impact of newly-established leadership hubs on nurses’ care of adults living with HIV/AIDS in Jamaica, Kenya, South Africa and Uganda was conducted between 2008 and 2012; HIV/AIDS is highly prevalent in all of these countries. Data were collected pre- and post- establishment of leadership hubs. We direct readers to Edwards et al. [20] for a description of analyses conducted to answer the primary study question, which was: What are the impacts of establishing “multi-stakeholder leadership hubs on evidence-informed HIV care practices?” [20]. Unfortunately, results indicated that leadership hubs lacked the force to enhance uptake of evidence-informed practice in HIV care. The authors concluded that hub success is dependent on greater integration within health authorities; such integration facilitates regularization of hubs and would enhance hub sustainability [20]. In this paper, using the original post-intervention sample with participants from all four countries, we examine the contribution of organizational interventions, including health care workplace policies, procedures and quality assurance initiatives, and self- and peer-assessed individual nurse practices, to nurse-reported HIV/AIDS-stigma practices toward patients living with HIV/AIDS and nurses in health care settings. The study questions were as follows:

Our hypotheses were that, in the study countries: (1) the presence of organization-level HIV/AIDS-related workplace policies and procedures would predict fewer incidents of nurse-reported HIV/AIDS stigma toward nurses and patients; (2) the presence of quality assurance initiatives would predict fewer incidents of nurse-reported HIV/AIDS stigma toward nurses and patients, and; (3) a higher quality of individual level nursing practices (whether self-assessed or peer-assessed) would predict fewer incidents of nurse-reported HIV/AIDS stigma toward nurses and patients.

For a comprehensive description of the leadership hub intervention, refer to Edwards et al. [20]. This intervention was implemented in three districts/parishes of each participating country—Jamaica, Kenya, South Africa and Uganda. Three control districts/parishes were also selected in each country, except South Africa, where no control districts were sampled. There were some country-specific differences noted at baseline (see Edwards et al. [20] for details); for example, Ugandan participants were significantly less likely to report the existence of quality assurance initiatives and/or evidence-based HIV/AIDS policies and procedures than those in other countries. Notably, quality assurance initiatives were less commonly reported than evidence-based HIV/AIDS policies and procedures in all four countries [20]. All publicly-run health care institutions meeting World Health Organization (WHO) classification criteria for a level 3, 4 or 5 institution were eligible to be included in the study [21]. Facility size ranged from small, sub-district level health centres (level 3), which primarily provide primary health care services, to large, provincial or national referral hospitals (level 5) [20, 21]. Institutional classification criteria were consistent across countries.

A random, stratified sample of health institutions was selected in each district. Each national, provincial and parish/district hospital was provided with the protocol to randomly sample staff for inclusion in the study. All eligible staff employed in health centres were asked to participate; criteria for eligibility included fluency in English, employment in the health centre for a minimum of 3 months, profession of registered or enrolled nurse, and employment as a staff nurse or manager. Although institutions sampled were the same both pre- and post- intervention, individuals sampled pre-intervention were not necessarily the same as those sampled post-intervention. Only one wave of data was required to answer our secondary question; we elected to use post-intervention data because they were collected more recently. As noted by Edwards et al. [20], there was a general trend toward fewer reports of nurse stigmatization of patients in the study countries between baseline and follow-up. This trend was observed in both intervention and control groups [20].

The baseline and follow-up questionnaires were identical. A complete description of the questionnaire is included in Edwards et al. [20]. For our analysis, we utilized post-intervention measures of clinical conduct, quality assurance initiatives, workplace policies and stigma. Measures of clinical conduct provided a self- and peer-assessment of the extent to which nurses or their peers consistently carried out evidence-based HIV/AIDS nursing practices. Measures of quality assurance and workplace policies examined nurses’ awareness of the existence of evidence-based quality assurance initiatives and of workplace policies in-line with evidence-based approaches to HIV/AIDS care. Measures were adapted from existing, validated instruments [22‐24].

Research assistants in each country entered and cleaned data using Microsoft Excel™. For more detail on the original quantitative analysis of the data, refer to Edwards et al. [20]. Prior to conducting our analysis, we calculated correlations using the Spearman method. The data, as collected, incorporated questionnaire responses from both managers and nurses working with these managers. Therefore, the nested structure of the data, with managers at the facility (hospital) level and nurses within each hospital at the individual level, allowed us to use HLM [28]. We were not able to link the data from each nurse to a specific unit or manager; therefore, we developed a two-level model (nurses at level 1 and managers and facility characteristics at level 2), rather than a three-level model with nurses at level 1, managers/units at level 2 and facilities at level 3. Next, we determined theoretically which variables should be situated at each level (as per measures reported above) and then calculated Intra-Class Correlations (ICC2 s), to evaluate whether data showed that nurses were more alike in their responses within their facility than across facilities. Generally, a value of 0.70 or more indicates sufficient agreement across individual responses to warrant aggregation to the group/facility level [29]. Due to the nature of our question—exploring the impact of institution level initiatives on HIV/AIDS-stigma and the inherent nesting in the data collected—HLM was the best method of analyzing these data.

To ensure that we had adequate power to conduct HLM we removed all institutions with fewer than 10 responses per institution. Fewer than 10 responses at an institution was not necessarily indicative of a poor response rate as sub-district and sub-parish health centres (Level 3 WHO Institutions) often had fewer than 10 nursing staff. This exclusion still provided the minimum sample of 25 facilities required at level 2 [30]. It is notable, however, that the number of Level 3 WHO institutions decreased to ten from 83 in the larger sample. Following this step, we reviewed the data to ensure that institution-level variables were defined for each included institution: as such we required a minimum of one manager response for each (see Fig. 1 for the analytic sample selection algorithm). Additionally, as HLM requires complete data (no missing values) for all variables of our model, we removed cases in which data were missing for any included variables. Thus, an observation was removed when a respondent did not answer one or more of the items in at least one of the survey sections. Scale values were derived as the average of all available survey items included in the scale (with reverse-coding as needed). See Fig.1 for details on the process of isolating the analytic sample, including 29 institutions and data from their managers and nurses.

We built the model step-wise by adding variables at level 1 first, then level 2, using model deviance statistic as a selection criterion. Predictor variables at the individual-level, utilizing responses from staff nurses, included self-assessment of clinical conduct (Clinical Conduct) and perceptions of coworkers’ clinical conduct (Co-Worker Clinical Conduct). At the facility level we included the measures of quality assurance initiatives (Quality Assurance) and workplace policies (Policies and Procedures) to determine whether having organizational workplace policies and procedures and quality assurance initiatives predicts HIV/AIDS-stigma behaviours toward nurses and patients. Manager responses within each organization were aggregated to the institutional level to represent the institution. Stigmatization—the outcome variable—was at the individual level, using data from staff nurses. Per this selection criteria, the variables described above were introduced into the model in the order of Clinical Conduct, Co-Worker Clinical Conduct, Policies and Procedures and Quality Assurance. The fitted models described, including a full univariate model that ignores the individual-facility hierarchical structure for comparison, are presented in Table 5.

The analytic sample (n = 1157) included managers (n = 392) and registered/enrolled nurses (n = 765) from 29 facilities—5 in South Africa, 10 in Uganda, 7 in Jamaica and 7 in Kenya. Ten of the facilities were health centers (Level 3 WHO Institutions), 8 (Level 4) were at the district/parish level and 11 were national/provincial (Level 5). Across the 29 facilities, a range of 1-88 and a median of 10 managers responded per institution. The managers in the sample were 89% female and 77% reported interacting with HIV/AIDS patients daily. Seventy percent of managers in the sample worked in national/provincial (Level 5) institutions, 18% in district/parish (Level 4) institutions and 12% in health centers (Level 3). The nurses in the sample were 93% female and 67% reported interacting with HIV/AIDS patients daily. See Table 1 and Fig. 1 for a description of the samples. See Table 2 for means, standard deviations (SDs), and ranges of scores for all included scales. No significant difference in scale means, SDs and ranges were seen between the full sample (including all participants) and those of the smaller, analytic sample (Table 3). Figure 2 visually displays inter-institution differences in average ordinal stigmatization score.

We calculated an ICC2 of 0.60 for our ordinal measure of stigma, supporting its use as a level 1 variable. See Table 2 for ICC1s. Absolute values for correlations between variables included in the model ranged from 0.001 to 0.836. See Table 4 for the correlation matrix. These correlation measures indicate a significant positive correlation between mean scores for clinical conduct and co-worker clinical conduct and between mean scores for clinical conduct and stigmatization. Significant positive correlation was also seen between measures of quality assurance and policies and procedures and between measures of policies and procedures and stigmatization. These correlation measures indicated significant negative relationships between clinical conduct and quality assurance and between both quality assurance and policies and procedures with co-worker clinical conduct.

The model testing results are presented in Table 5. In the final model (Model 4), both clinical conduct and co-worker conduct significantly predicted stigmatization, with higher (better) self-assessed scores in clinical conduct significantly predicting higher (worse) stigmatization scores of people with HIV/AIDS, and higher (better) scores for perceptions of co-worker clinical conduct significantly predicting lower (better) stigmatization of people with HIV/AIDS. Despite not presenting as a significant predictor of stigmatization (p = 0.11), the addition of the institution-level measure of policies and procedures contributed 12% explained variance in stigmatization and, when added, significantly improved model fit.

In our study, our analysis may have yielded more specific results if the tool used to evaluate HIV/AIDS stigma and discrimination had elicited more varied responses from study participants. Issues with measures of stigma have been reported by both Stangl et al. [10], who noted that existing scales have not been validated across multiple populations and contexts and Sengupta et al. [35], who concluded that many scales focus on assessing stigma in groups of uninfected individuals rather than in mixed populations (inclusive of infected and uninfected individuals). The stigma scale utilized in this study, however, was subjected to psychometric testing across five countries [25]. Stigma scores are often influenced by social desirability bias [36], which leads respondents to model their responses on cultural mores rather than on their beliefs and experiences. In this case, respondents may have considered a response at the lower end of the scale (i.e. never having observed stigma toward patients by nurses or stigma toward nurses) to be more socially desirable than a response at the high end of the scale (reflecting observed instances of stigma toward patients by nurses or toward nurses that occur “most of the time”). Social desirability is particularly troublesome in research designs that incorporate pre- and post-test assessments as participants may infer that researchers desire a reduction in the measure (in this case stigma) over time [36]. The lack of variance led us to merge the two dimensions of HIV/AIDS stigma (i.e. stigma by nurses against people with HIV/AIDS and stigmatization of nurses by co-workers and community members) and to collapse categories of frequency. Advancements in the field of stigma measurement may allow researchers to better differentiate between the relative effectiveness of workplace policies and quality assurance initiatives (and other interventions) to reduce stigma, as experienced by different targets.

As noted above, a limitation of HLM was the requirement for a minimum of 10 individual-level responses per institution; for this reason many institutions (with fewer than 10 individual-level responses) had to be removed from the analytic sample. Additionally, we were unable to directly connect an employee to their manager, which prevented us from conducting a more complex analysis that could have accounted for variance at three levels (staff, manager and facility). The role of manager likely varied across facilities—managers in WHO level three facilities would be more likely to be involved in direct clinical care while those in WHO level 5 facilities may have been directly involved in crafting policies and procedures and in designing and implementing quality assurance initiatives. In future studies, where feasible, information on reporting structure could be recorded. This would allow us to conduct analyses that could parse out the role of managers in reducing HIV/AIDS stigma.