The impact of an unknown HIV serostatus on inpatient mortality

Introduction Determining HIV serostatus is crucial for linking HIV-infected patients to appropriate care, which might reduce their risk of subsequent morbidity and mortality. A recent South African study demonstrated a potentially harmful association between an unknown HIV serostatus and rehospitalisation. The impact of an unknown HIV status on inpatient mortality has not yet been established in that setting, which formed the impetus for the current study. Methods This was an unmatched case-control analysis of adult patient data collected as part of a demographic survey at the Hlabisa Hospital, South Africa between October 2009 and February 2014. Cases were defined as patients who suffered inpatient mortality, while controls were patients who did not suffer inpatient mortality. A sample size of 92 cases and 276 controls was used in this study. Patient data related to age, gender, distance between referral clinic and the hospital, HIV serostatus (HIV-negative, HIV-positive or an unknown HIV serostatus) and comorbidity were analysed using recommended methods for unmatched case-control studies. Results When potential confounders were accounted for, we found an unknown HIV serostatus to be associated with an almost 8-fold increase in the odds of inpatient mortality when compared with patients who were known HIV-negative (Odds Ratio: 7.64, 95% Confidence Interval: 1.11-52.33, p = 0.038). Conclusion An unknown HIV serostatus was independently associated with a higher odds of inpatient mortality. This finding highlights the potential benefit of adopting an “opt-out” approach to HIV counseling and testing. Further research on this topic is required.


Introduction
The Southern African region continues to be disproportionately burdened by HIV/AIDS. South Africa is one of the most severely afflicted countries in this region [1]. Despite the estimated prevalence of HIV in the population ranging between 11% and 20%, half of all adult South Africans do not known their HIV serostatus [2]. Determining HIV serostatus is crucial for linking HIVinfected patients to appropriate care, which might reduce their risk of subsequent morbidity and mortality [3,4]. Large clinical studies investigating patient outcomes (including mortality) are scarce in resource-limited settings due to logistical challenges in creating and maintaining patient registries at healthcare facilities [5]. This has since improved with the establishment of surveillance mechanisms, such as the publically-available Africa Centre Demographic Information System (ACDIS), which also collected inpatient data from a hospital located in a South African setting with a high prevalence of HIV infection [6]. A recent analysis of this hospital data collected as part of ACDIS suggested a statistical trend toward a potentially harmful association between an unknown HIV serostatus and rehospitalisation [7]. The impact of an unknown HIV serostatus on inpatient mortality in the same setting is yet to be established. An understanding of any potential association between an unknown HIV serostatus and inpatient mortality might have important implications on initiatives aimed at improving the uptake of HIV counseling and testing in this setting, as well as improving subsequent linkage to care in patients with previously undiagnosed HIV infection. This formed the impetus for the current study.

Methods
Study design, setting and study population: We performed an unmatched case-control analysis of adult medical inpatient data collected between October 2009 and February 2014 at the Hlabisa Hospital in KwaZulu-Natal Province, South Africa as part of ACDIS [6]. The hospital is considered an important primary care facility in the region, serving a predominantly rural population. HIV incidence in the Hlabisa district of KwaZulu-Natal has been estimated at 3.2/100 person years [6]. The methodology suggested by Breslow et al. was used to conduct this unmatched case-control study [8].
Case and control definitions: Cases were defined as patients who suffered mortality while hospitalized. Inpatient mortality was determined from the patient discharge disposition variable in the ACDIS Hlabisa Hospital dataset. Controls were defined as patients who did not suffer mortality while hospitalized.

Sample size calculation:
The sample size required for this study was 368 (92 cases and 276 controls) patients. The following parameters were used to calculate the study sample size: anticipated odds ratio -2.0, estimated exposure of controls -25% (anticipated to be lower than the 50% reported for the general South African adult population as it was likely that patient HIV serostatus might have been determined at other healthcare facilities), alpha risk -5%, power -80%, and a case:control ratio of hypertension, diabetes, cancer, cardiovascular disease (a composite of myocardial infarction, stroke and heart failure) and renal disease [9]. ICD-10 codes for tuberculosis and pneumonia were adopted from those used by the Healthcare Cost and Utilization Project [10].
We also considered that the distance between the referral clinic and the hospital might also be an important factor impacting inpatient outcomes and we also included this variable in our final data analysis.
Statistical analysis: Case-control data were initially analyzed using X 2 , Fisher's Exact or Mann-Whitney tests where appropriate.
Results for the aforementioned crude data analyses are presented as frequencies with percentages or medians with interquartile ranges (IQR). An unconditional logistic regression model was used to account for potential confounding in the unmatched case-control study design, with the results of this adjusted statistical analysis being presented as odds ratios (OR) with 95% confidence intervals (95% CI). A p-value of <0.05 was considered statistically

Results
Descriptive statistics and crude analyses: Results of the descriptive and crude statistical analyses are presented in Table 1. Secondly, there need to be measures in place to ensure that facilities have access to HIV testing kits and no kit stock outs occur.
Thirdly, care structures to which newly diagnosed patients are referred must be improved and prepared for an increase in patient volumes. Importantly, these challenges will result in an increase in healthcare expenditure and would require careful allocation of healthcare resources. Careful planning and consultation amongst public health specialists would be required to overcome these potential challenges for an "opt-out" approach to HIV counseling and testing to be viable in our setting. While we found several other patient/clinical characteristics to be independently associated with/not independently associated with a higher odds of inpatient mortality, these findings should be interpreted with caution. The sample size used in this study was derived with specific reference to determining the impact of HIV serostatus on inpatient mortality, while other characteristics were included to merely adjust for potential confounding during the unconditional logistic regression analysis. Therefore, it is likely that this study was not adequately powered to investigate the impact of several of the other patient/clinical characteristics mentioned above. Further casecontrol analyses (with appropriately revised sample sizes) for each patient/clinical characteristic of interest would be required to draw conclusions related to the impact of these patient/clinical characteristics on inpatient mortality. There were additional limitations to our study. Firstly, this study was an analysis of data from a single rural healthcare facility. This entails that the findings of this study might not necessarily be generalizable to other healthcare facilities. A larger multicenter study would be required to confirm our findings. Secondly, only inpatient mortality was investigated in this research. While 30-day mortality would be a more appropriate outcome, patient loss to follow-up is a major challenge in resource-limited settings [14]. Strategies for estimating 30-day mortality in medical patients in these settings are required.
Nevertheless, our use of inpatient mortality as our study outcome is justified as this outcome is an important outcome used in healthcare facilities [15]. Lastly, data related to medication use was not collected as part of the hospital database and we were unable to adjust for the effects of medication use in our logistic regression model. Prospective research is required to investigate the potential impact of medication use on inpatient mortality in our setting.

Conclusion
In conclusion, we found that when compared with a HIV-negative serostatus, an unknown HIV serostatus was associated with an almost 8-fold increase in the odds of suffering inpatient mortality in our setting. It is possible that a shift toward a systematic "opt-out" approach to HIV counseling and testing in our setting might be useful in identifying undiagnosed HIV-positive patients and initiating these patients on antiretroviral therapy for improved survival.
However, this would require careful planning to ensure its viability.
Further research is required to confirm our study findings.
What is known about this topic  Large clinical studies investigating HIV patient outcomes (including inpatient mortality) are scarce in resourcelimited settings;