Validation of the Data Collection on Adverse Events of Anti-HIV Drugs (D:A:D) chronic kidney disease risk score in HIV-infected patients in the USA

Abstract

Objectives: The aim of the study was to assess the validity of an easy-to-calculate chronic kidney disease (CKD) risk score developed by the Data Collection on Adverse Events of Anti-HIV Drugs (D:A:D) group in a longitudinal observational study of people living with HIV (PLWH) in the USA.

Methods: PLWH (2002-2016) without prior exposure to potentially nephrotoxic antiretroviral agents and with at least three estimated glomerular filtration rate (eGFR) test results were identified in the Observational Pharmaco-Epidemiology Research and Analysis (OPERA^® ) cohort. Three samples were drawn independently using the same eligibility criteria but each using a different eGFR equation, specifically the Cockcroft-Gault (C-G), Modification of Diet in Renal Disease (MDRD) or Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) eGFR estimation method. Full and short D:A:D risk scores were applied. CKD was defined as a confirmed decrease in eGFR to < 60 mL/min/1.73 m² (stages 3-5). Poisson models estimated the association between CKD incidence and a one-point increase in the continuous risk score. The incidence rate ratio (IRR), adjusted IRR (aIRR), and Harrell's discrimination statistic were used to assess validity.

Results: There were 19 444, 22 727 and 22 748 PLWH in the OPERA C-G, CKD-EPI and MDRD samples, respectively. The median (minimum-maximum) follow-up duration was 6.1 (0.3-9.1) years in the D:A:D cohort and ranged from 3.2 to 3.5 (0.2-15.5) years in the OPERA validation samples. The observation time for the majority of PLWH in the D:A:D cohort began prior to 2006, in stark contrast to the OPERA validation samples, where the majority of PLWH were observed after 2011. The CKD incidence ranged from 7.3 per 1000 person-years [95% confidence interval (CI) 6.8, 7.9 per 1000 person-years] in OPERA C-G to 11.0 (95% CI 10.4, 11.6 per 1000 person-years) in OPERA MDRD. In OPERA samples, IRRs by risk group and adjusted IRRs (full risk score) were similar to those in the D:A:D derivation cohort (adjusted IRR 1.3; 95% CI 1.3, 1.3). Harrell's c-statistic ranged from 0.87 to 0.92 in the OPERA samples, comparable to that in the derivation cohort (0.92). Results for short scores were similar.

Conclusions: The findings support the validity of the D:A:D risk scoring method for assessing CKD (stages 3-5) probability in an exclusively USA-based sample regardless of eGFR method.

Keywords: Data Collection on Adverse Events of Anti-HIV Drugs (D:A:D) risk score; HIV; chronic kidney disease; renal impairment; validation.

Figure 1

People living with HIV in the Observational Pharmaco‐Epidemiology Research and Analysis (OPERA) database and Centers for Disease Control and Prevention (CDC) (2010) state‐by‐state estimates of HIV infection in the USA.

Figure 2

Baseline patient characteristics and Data Collection on Adverse Events of Anti‐HIV Drugs (D:A:D) full and short risk calculations, by Observational Pharmaco‐Epidemiology Research and Analysis (OPERA) validation cohort. ART, antiretroviral therapy; eGFR, estimated glomerular filtration rate; NA, not applicable.

Figure 3

Crude chronic kidney disease (CKD) incidence and 95% confidence intervals, by study population and CKD full risk score. *The scale is logarthmic.

Figure 4

Incidence rate ratios and 95% confidence intervals for the association between chronic kidney disease (CKD) development and CKD full risk group, with medium risk as the referent. *The scale is logarthmic.

Figure 5

Adjusted incidence rate ratio and 95% confidence intervals for the association between chronic kidney disease (CKD) development and a one‐point increase in the continuous CKD full risk score. *Modelling in the Data Collection on Adverse Events of Anti‐HIV Drugs (D:A:D) derivation cohort was adjusted for baseline injecting drug use, gender, hepatitis C virus coinfection, age, nadir CD4 count, estimated glomerular filtration rate (eGFR), hypertension, prior cardiovascular disease (CVD) and diabetes. Models in all other cohorts were not further adjusted.