Chronic kidney disease in the global adult HIV-infected population: A systematic review and meta-analysis

Introduction The widespread use of antiretroviral therapies (ART) has increased life expectancy in HIV patients, predisposing them to chronic non-communicable diseases including Chronic Kidney Disease (CKD). We performed a systematic review and meta-analysis (PROSPERO registration number CRD42016036246) to determine the global and regional prevalence of CKD in HIV patients. Methods We searched PubMed, Web of Science, EBSCO and AJOL for articles published between January 1982 and May 2016. CKD was defined as estimated glomerular filtration rate (eGFR) <60ml/min using the MDRD, Cockcroft-Gault or CKD-EPI equations. Random effects model was used to combine prevalence estimates from across studies after variance stabilization via Freeman–Tukey transformation. Result Sixty-one eligible articles (n = 209,078 HIV patients) in 60 countries were selected. The overall CKD prevalence was 6.4% (95%CI 5.2–7.7%) with MDRD, 4.8% (95%CI 2.9–7.1%) with CKD-EPI and 12.3% (95%CI 8.4–16.7%) with Cockcroft–Gault; p = 0.003 for difference across estimators. Sub-group analysis identified differences in prevalence by WHO region with Africa having the highest MDRD-based prevalence at 7.9% (95%CI 5.2–11.1%). Within Africa, the pooled MDRD-based prevalence was highest in West Africa [14.6% (95%CI 9.9–20.0%)] and lowest in Southern Africa (3.2%, 95%CI 3.0–3.4%). The heterogeneity observed could be explained by WHO region, comorbid hypertension and diabetes mellitus, but not by gender, hepatitis B or C coinfection, CD4 count or antiretroviral status. Conclusion CKD is common in HIV-infected people, particularly in Africa. HIV treatment programs need to intensify screening for CKD with added need to introduce global guidelines for CKD identification and treatment in HIV positive patients.


Introduction
The widespread use of antiretroviral therapies (ART) has increased life expectancy in HIV patients, predisposing them to chronic non-communicable diseases including Chronic Kidney Disease (CKD). We performed a systematic review and meta-analysis (PROSPERO registration number CRD42016036246) to determine the global and regional prevalence of CKD in HIV patients.

Methods
We searched PubMed, Web of Science, EBSCO and AJOL for articles published between January 1982 and May 2016. CKD was defined as estimated glomerular filtration rate (eGFR) <60ml/min using the MDRD, Cockcroft-Gault or CKD-EPI equations. Random effects model was used to combine prevalence estimates from across studies after variance stabilization via Freeman-Tukey transformation.

Introduction
Chronic Kidney Disease (CKD) is a worldwide public health problem; moving from 27 th to the 18 th most important global cause of death within the last 2 decades [1]. This degree of shift was second only to HIV/AIDS [1], suggesting a significant relationship between HIV and CKD as an important intersection between chronic non-communicable diseases (NCDs) and communicable diseases.
With the roll-out of antiretroviral therapies (ARTs), individuals with HIV are now living longer. As a consequence, the spectrum of kidney diseases in HIV patients has broadened, ranging from asymptomatic changes in renal function like proteinuria, [2,3] electrolyte losses [4] and acute kidney injury [5] occurring from diarrheal illnesses to various degrees of CKD occurring as a result of renal damage from chronic non-communicable diseases or HIV-associated nephropathy (HIVAN). Furthermore, the use of certain medications included in some ART regimens such as tenofovir and ritonavir, has been shown to increase the risk of CKD [6]. Among incident end-stage renal disease (ESRD) patients, HIV has been implicated as the etiologic factor in 0.4%-0.7% of patients in France [7,8]; 0.5%-1.1% in Spain [9, 10]; 6.6% in Cameroon [11]; and 28.5% in South Africa [12]. One large study has shown that as much as 3.3% of HIV positive patients with normal baseline estimated glomerular filtration rate (eGFR) developed CKD over a relatively short follow up period of 3.7 years, highlighting the burden of kidney disease in HIV patients [13]. The prevalence of CKD in HIV-infected individuals varies widely between geographic regions and depends on the reporting methods and the definition of CKD used, ranging from 2% to 38% [14,15]. Although there is an increasing number of individual reports on the prevalence of CKD in the HIV population, the data have not been appropriately synthesized to date.
In this analysis, we synthesized available data on CKD prevalence in the adult HIV population at both regional and global levels. The overarching goal was to provide an essential basis to guide contextualized effective prevention and control strategies to tackle the burden of CKD in this population.

Selection of studies for inclusion in the review
The Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) 2009 guidelines [16] served as the template for reporting the present review (S1 Table, Fig 1). The study protocol was published at the International Prospective Register of systematic reviews, (PROSPERO registration number CRD42016036246). All observational studies and clinical trials reporting on the prevalence of CKD in HIV-infected adults (! 18 years) or providing enough data to compute it, using established creatinine-based equations [Modification of Diet in Renal Disease (MDRD) [17], Cockcroft-Gault (CG) [18], Chronic Kidney Disease Epidemiology (CKD-EPI) [19]] to estimate GFR were included. CKD was defined as eGFR <60ml/ min/1.73m 2 irrespective of proteinuria status. Studies that reported CKD as eGFR <60ml/ min/1.73m 2 and/or persistent proteinuria were only included if we could compute the frequency of those with eGFR <60 ml/min/1.73m 2 from available data in the article. We also included studies that reported CKD prevalence using a single estimated eGFR in order to accommodate studies from low-income countries where repeated serum creatinine measurement might not have been performed. A comparison of the pooled prevalence from studies with a single eGFR estimate and that with multiple estimates was also undertaken. We excluded studies with small sample size (<100 participants) and those including both adult and pediatric populations in which it was not possible to disaggregate data for adults. For studies published in more than one report (duplicates), the most comprehensive reporting the largest sample size was considered.

Identification of studies
We searched PubMed/MEDLINE, EBSCO, Web of Science and African Journals Online to identify all relevant articles reporting data on the prevalence of CKD in HIV-infected adults published from January 1, 1982 (when the HIV epidemic started) to September 30, 2016. We conceived and applied a search strategy based on the combination of relevant terms relating to HIV and CKD. The search strategy for Pubmed, web of science, EBSCO and AJOL is shown in S2 Table. No language restrictions were applied. References of all relevant research articles and reviews were also scrutinized to identify additional potential data sources.

Assessment of methodological quality of included articles
The methodological quality of included studies was evaluated using the 9-point rating system developed by Stanifer et al [20] and modified for the purposes of this study. The scoring criteria for quality of studies is shown in S3 Table while S4 Table shows the methodological quality of the included articles. The scoring criteria assessed factors related to representativeness of the study participants, sampling, sample size and assessment of possible confounders to the relationship between HIV and CKD. Studies were rated as having a high, medium or low methodological quality when they were assigned a score higher than 6, 5 and 6; or 4 and below respectively.

Study selection and data extraction
Two investigators (UEE and IGO) independently screened the titles and abstracts of articles retrieved from literature search, and the full-texts of articles found potentially eligible were obtained and further assessed for final inclusion (Fig 1). Disagreements were resolved by consensus or consultation of a third investigator (APK). For clinical trials, we used baseline data. A World Health Organization (WHO) region [21] was assigned to each study depending on the country of recruitment. All studies from Africa were subsequently seperated from the rest and further sub-divided into the different African Union (AU) sub-regions [22] for the purpose of statistical comparisons. Two investigators (UEE and EEE) independently extracted data; discrepancies between investigators were resolved through discussion until consensus was achieved. In one instance [23], an author was contacted for clarification where data was uncertain. Data extracted included first author name, year of publication, country of study origin, WHO region, African sub-region (if study was from Africa), gender proportions in the study population, median age, Body mass index (BMI), CD4 count and viral load of the study population, prevalence of hepatitis B, C co-infection; and the prevalence of hypertension and diabetes mellitus in the study population.

Statistical analyses
A meta-analysis was used to summarize prevalence data. We pooled the study-specific estimates using a random-effects meta-analysis model (DerSimonian-Laird) to obtain an overall summary estimate of the prevalence of CKD according to the different eGFR equations across studies [24], after stabilizing the variance of individual studies with the use of the Freeman-Tukey double arcsine transformation to minimize the effect of extreme prevalence on the overall estimate [25]. Heterogeneity was assessed using the χ 2 test on Cochrane's Q statistic [26] and quantified by calculating the I 2 (with values of 25%, 50% and 75% representing low, medium and high heterogeneity respectively) [27]. Subgroups analysis was also performed using the Q-test based on ANOVA. We assessed the presence of publication bias using funnel plots and the Egger's test [28]. We assessed inter-rater agreement for study inclusion and data extraction using Cohen's kappa (κ) coefficient [29]. A p-value <0.05 was considered indicative of statistically significant difference between subgroups. Data was analyzed using the statistical software Open Meta Analyst

Results
The initial literature search retrieved 1220 articles of which 99 were selected after title and abstract screening for full-text review. Finally, 61 articles [23, 32-91] were eligible and included in this systematic review (Fig 1). There was a high agreement between investigators for study inclusion (κ = 0.81).
Using the MDRD equation, the African region had the highest prevalence estimate at 7.9% (95%CI 5.2%-11.2%) while the European region had the lowest estimate at 3.7% (95%CI 2.5-  Table 2; summary statistics from meta-analyses of prevalence studies on CKD in people with HIV using random effects model and arcsine transformations are shown in S5 Table. CKD prevalence was also high in Africa using the CKD-EPI equation: 7.0% (95%CI 2.8-12.9%). Studies using CG equation were mostly from Africa (84% of the studies), precluding sound regional analysis. The pooled prevalence of CKD in Africa from CG estimator was 13.7% (95%CI 9.1-19.0%); Table 2.

Discussion
To our knowledge, this is the first attempt to provide prevalence estimates for CKD in HIV populations across various WHO regions. Prevalence was highest in Africa and lowest in Europe although the data shows substantial heterogeneity. Despite this, sociodemographic and clinical factors such as gender, age, coinfections with HBV and HCV did not significantly affect the estimates while coincident hypertension and diabetes mellitus had significant effect on the estimates. Paradoxically, our analysis did not reveal significant contribution to CKD prevalence of HIV related factors such as CD4 counts and ART status.
The overall prevalence of CKD in HIV populations is high, regardless of estimator used. This is more so in Africa where the prevalence of CKD in the general population is already high [20]. The high CKD prevalence in HIV patients presents an enormous challenge to health care systems in low to middle income countries (LMICs) with high prevalence of HIV and where access to CKD care is significantly lacking [92]. The clinical and economic implication of a high CKD burden has effects on the functioning of health systems. In higher income countries, high CKD burden may represent remarkable increase in healthcare costs for managing HIV related CKD whereas in LMICs, it may mean enormous pressure on an already weakened and poorly funded health system. The interplay between HIV and CKD also presents an opportunity for integration of chronic non-communicable disease care with communicable disease treatment as this may enhance more effective use of health resources and improve long term outcomes for HIV patients. It is important to determine if there is a higher CKD prevalence among HIV populations than the general population. In Africa, the prevalence of Prevalence of CKD in the global HIV population individuals with eGFR less than 60ml/min/1.73m 2 in the general population is not clear but Stanifer et al [20] reported a pooled prevalence of 13.9% using both eGFR and proteinuria in the definition of CKD. Studies in Sub-saharan Africa have reported a prevalence of eGFR less than 60ml/min/1.73m 2 of 1.6% [93] to 8.0% [94] using the MDRD formula. In this analysis, the pooled prevalence (using MDRD) was 7.9% for Sub-Saharan Africa. However, studies undertaking head-to-head comparison of CKD prevalence in the HIV-infected population and the general population [95,96] in climes with better data collection suggest higher CKD prevalence in the HIV population than the general population. Consistently, there was significant difference in the prevalence reported across the three estimators. Prevalence estimates obtained using the CG equation were generally higher than those obtained from MDRD and CKD-EPI with CKD-EPI being the most conservative of the three. In the general population, the CKD-EPI equation appears to outperform the MDRD and CG equations [97][98][99]; however, the best equation for GFR estimation and cut-off for definition of CKD in HIV patients remains controversial [100]. Some authors have suggested that existing equations do not take into account the lean muscle mass of malnourished HIV patients and the lipodystrophy associated with ART use [101]. One study report suggests that the CKD-EPI equation may underestimate CKD prevalence in the HIV population in Africans [102]. Whether this is also applicable to European or North American HIV populations, is uncertain. Other studies have supported the idea that eGFR values obtained from CG do not have clinically significant difference from those obtained from CKD-EPI equation in HIV patients and so could be used interchangeably [103], while MDRD is thought to be less sensitive to moderate GFR reductions and thus not useful in HIV patients with early CKD [64]. Noteworthy is the observation that most of the studies from Africa (where patients present with advanced HIV disease) used the Cockroft-Gault equation to estimate GFR either alone or in combination with other creatinine-based formulae. This may be responsible for the relatively higher CKD prevalence obtained from the Cockroft-Gault equation. There have been attempts at validating these creatinine-based estimators in the HIV population [104][105][106] but there is yet no consensus on the best creatinine-based GFR estimator in this special population.
Although not statistically significant in most of the comparisons, there was clearly a trend towards lower CKD prevalence estimates in the studies with more than one GFR estimate compared with those with only one estimate. This validates the KDIGO position of demonstration of GFR <60ml/min/1.73m 2 for at least 3 months [107] before a firm diagnosis of CKD is made. This may provide evidence of significant risk of overestimation of CKD prevalence in single eGFR studies because of the possibility of undiagnosed acute kidney injury (AKI) especially in patients with HIV who tend to have higher risk of AKI than the general population [108].
Hypertension and diabetes mellitus remains significant risk factors for CKD in the HIV population as seen in this analysis when head-to-head comparison was performed between HIV only cohorts and HIV/hypertension or DM co-morbidities. Both hypertension and diabetes mellitus are age-related conditions and with the increasing age of HIV patients, a higher prevalence of CKD might be predicted in future in HIV positive patients. Both conditions, however, did not appear to explain some of the heterogeneities in CKD prevalence estimates, when comparison was made based on median hypertension or DM prevalence (S5 Table). This may not be unconnected with the lack of uniformity in the definition or method of assessment of these factors among the constituent studies.  90] or inadequate information about criteria for diagnosis [82]. However, multivariate regression in some of the component studies [38,82,109] identified significant association between diabetes and hypertension with CKD in HIV patients.
The effect of hepatitis B and/or hepatitis C on CKD occurrence in HIV patients has not been consistent. In this study, we found no significant difference in the pooled prevalence of studies with high hepatitis B or C co-infection compared with those with low prevalence of these viral co-infections. Some observational studies have found a higher risk of CKD [35, 90,110] among hepatitis B or C co-infected HIV patients while others found no significant effect with hepatitis B or C co-infection [36,37]. A meta-analysis investigating the effect of hepatitis C co-infection on CKD occurrence and progression in HIV patients [111] found significantly increased risk of CKD, proteinuria and AKI in co-infected individuals compared to those with only HIV infection. We are unaware of any published meta-analysis comparing CKD prevalence or progression in HIV-hepatitis B co-infected individuals with HIV only patients though observational studies [110,112] suggest increased CKD risk with hepatitis B co-infection. Aggregation of data from high-income countries (high HCV co-infection and relatively low CKD prevalence) with LMIC (low HCV co-infection and high CKD-HIV prevalence) may have led to a loss of significant difference in CKD prevalence in the HIV-HCV co-infected compared to those without the co-infection.
One possible reason for the relatively high prevalence of CKD in African patients is late presentation to HIV care clinics at advanced stages of disease. This is evidenced by the significantly lower CD4 counts in African patients compared to the other regions. This may be compounded by late initiation of anti-retroviral medications giving adequate time for HIVinduced or related damage to the kidneys. In North America, ARTs are given to all HIVinfected individuals regardless of CD4 count to reduce morbidity and mortality associated with HIV infection [113]. This is has not been the case in most SSA countries where cut-offs of CD4 counts were used for initiation of ART [South Africa (2013), < 350cells/μl [114]; South Africa (2015), <500 cells/μl [115]; Nigeria (2007), <200/μl [116]; Nigeria (2010), <350 cells/μl [117]]. It was only in 2016 that ART initiation was done regardless of CD4 count in some SSA countries. The effect of this policy change on CKD prevalence among HIV patients may only become apparent in the future. Early initiation of ARTs, especially in blacks, has been proposed as one of the measures for preventing CKD progression among HIV patients [118]. As more patients in SSA access ARTs it is possible that the incidence of CKD may not be too different between individuals of SSA origin compared with Caucasians [81]. There is also the problem of poor and inadequate facilities for long term monitoring of HIV patients on ARTs in Africa which makes early diagnosis of CKD difficult. Furthermore, CKD in HIV patients may occur because of repeated episodes of undocumented AKI. AKI is common among HIV patients and is an important cause of morbidity and mortality in this patient group with sepsis and hypovolemia from diarrhea being the commonest causes [119][120][121]. AKI has also been documented as an independent risk factor for future ESRD with increasing ESRD risk associated with worsening AKI stage in HIV patients [109,122].
The higher prevalence of HIV-related kidney disease in African Americans compared to Caucasian Americans [52] and very high CKD prevalence among HIV patients in West Africa suggests a possible genetic role in the increased CKD prevalence in SSA. This hypothesis is further strengthened by the observation that most African Americans are of West African origin and this study has shown the highest prevalence of CKD in HIV among West Africans. APOL1 and MYH9 polymorphisms have been implicated in conferring possible increased risk of CKD in Africans [123][124][125][126][127] but there may be more, yet to be identified, genetic risk factors. There also may be confounding environmental factors in Africa contributing to the increased CKD risk among HIV patients.
The global HIV population is quite heterogeneous; male preponderance in North America and Europe whereas females constitute 60-70% of the HIV patients in the African studies reviewed. The influence, if any, of gender difference on the CKD prevalence remains unclear. The prevalence of traditional risk factors for CKD like hypertension, diabetes mellitus and Hepatitis C is also higher in North America and Europe than in Africa. The high prevalence of these modifiable CKD risk factors present a window of opportunity for sustaining therapies that may ultimately slow down CKD progression. The experience garnered from chronic care management of HIV could be leveraged as a platform for integration of non-communicable disease services into HIV populations. The different dimensions of HIV care-prevention, diagnosis, enrollment into care, disease management and palliative care-could also be useful for NCDs. The integrated care model appears to have achieved good results in parts of SSA [128] and emphasis on CKD preventive services among the HIV population may reduce the burden of CKD in LMICs.
There is still inadequate information about the best creatinine-based eGFR formula for Africa in general [93] and the HIV population specifically and as our study has not been able to address this, it is a limitation. Some have suggested that non-inclusion of race to the MDRD equation may improve eGFR estimation in Africans [129] but this has not been validated in the HIV population. The use of Cystatin C is not yet widespread in Africa and may not be sustainable in Africa because of the cost. It is important to determine the best measure of CKD in this special population. We did not include individuals with eGFR greater than 60mls/min/1.73m 2 and persistent proteinuria in this study. If the definition of CKD was made to include persistent proteinuria, then the prevalence of CKD among HIV patients may be much higher than reported in this study. The lack of information on specific antiretroviral drugs and their potential contribution to the burden of CKD in this work is a limitation.
The burden of CKD in HIV positive patients is high globally, particularly in African patients. HIV treatment programs need to intensify routine screening for CKD at baseline and ART follow up clinics using relatively cheap and simple test for urinary proteins. There is now a great need to produce global guidelines for CKD identification and treatment in HIV patients and integrate treatment for chronic non-communicable disease with HIV patient care.
Supporting information S1