Session Type: Poster Session D
Session Time: 8:30AM-10:30AM
Background/Purpose: HCQ is an antimalarial drug effective in the treatment of rheumatologic conditions. High blood levels, advanced age, and extended treatment duration are associated with HCQ-induced retinopathy (RHCQ)(Petri et al. [Arthritis Rheum] 2020). We combined these insights with large amounts clinical therapeutic drug monitoring (TDM) data and deidentified patient information to develop interpretable linear models for the estimation of the overall prevalence of RHCQ in a cohort of HCQ-therapy-compliant patients, and the odds of an individual patient presenting RHCQ given HCQ dose and age.
Methods: US ordering providers (n=986) submitted specimens collected from patients receiving therapy HCQ to Exagen Inc’s CAP accredited clinical laboratory between December 2014 and April 2021 (Table 1). HCQ levels were measured in EDTA blood or capillary blood collected on micro-samplers using LC-MS/MS and reported to clinicians within 5 days of specimen receipt. Recommendation was made to collect specimens after 6 months therapy (steady state). The relationship between age per 10 years and HCQ dose per 200 mg qd with the likelihood of having HCQ levels greater than 1177 ng/mL was modeled using random intercept logistic regression. That estimated likelihood was combined with estimates of the sensitivity and positive predictive value of the cutoff HCQ ≥ 1177 ng/mL for the prediction of the incidence of RHCQ; thus allowing the estimation of the likelihood of RHCQ given age and HCQ dose and the estimation of the overall rate of RHCQ. Probabilistic modeling was performed with PyMC3 and compared to generalized estimating equations (GEE) implemented with StatsModels. Random intercept and GEE grouping was performed per ordering provider using only the most recent visit per patient. To ensure the analysis was performed using patients likely to be at least partially adherent to HCQ therapy, patients having HCQ < 500 ng/mL were excluded.
Results: Odds of having HCQ ≥ 1177 increased by 22% (CI95%Proba: [17%, 27%] | CI95%GEE: [15%, 26%]) per 10 years of age and 158% (CI95%Proba: [129%, 189%] | CI95%GEE: [102%, 206%]) per 200 mg HCQ prescribed qd. Figure 2 presents the theoretical population mean estimates (regression estimates with CI95%) and prediction intervals for individuals drawn from the cohort in increments of 10% from 10%-90%, and 95%. The estimated incidence of RHCQ in the cohort of patients having HCQ ≥ 500 was 5.72% (CI95% = [3.91%, 7.78%], Figure 3).
Conclusion: The estimated CI95% ([3.91%,7.78%]) for the prevalence of RHCQ in patients compliant to HCQ-therapy overlapped with estimates produced by previous studies. Model estimates for the population mean likelihoods agreed with individual prediction intervals, but the relatively wider prediction intervals emphasize the importance of appropriate utilization of HCQ TDM in patients chronically prescribed HCQ.
Table 1: Patient Characteristics. ‘Patients with HCQ Dose’ refers to the subset of patients for whom HCQ monitoring was ordered, and the ordering provider opted to indicate dosing information at the time of test requisition submission. The subset ‘Patients with HCQ Dose’ was further filtered to include only patients with venous blood HCQ ≥ 500. ‘All Patients with HCQ Order’ refers to any patient for whom HCQ monitoring was ordered, including non-compliant patients and patients for whom HCQ dose was not indicated. Only the most recent visit per patient are included.
Y-Axes are shared across each row; x-axes are conserved across all subplots. Top: Theoretical population mean estimates for the likelihood of achieving HCQ≥1177 ng/mL in a cohort of HCQ-therapy-compliant patients with prediction intervals describing the estimated likelihood of HCQ≥1177 ng/mL for individuals drawn from a population described by our cohort in increments of 10%, for 10%-90%, and 95%. Middle: Theoretical population mean estimates for the likelihood of R_HCQ in a cohort of HCQ-therapy-compliant patients with prediction intervals describing the estimated likelihood of R_HCQ for individuals drawn from a population described by our cohort in increments of 10%, for 10%-90%, and 95%. Bottom: Histograms describing the distribution patients by age allow an examination of the weight assigned to each vertical slice of likelihood in the middle plot when constructing the estimate of R_HCQ in Figure 3.
Top: Kernel density plot of estimated R_HCQ shows the uncertainty in the estimate of R_HCQ in a cohort of HCQ-therapy-compliant patients with CI95% and mean R_HCQ. Bottom: Box plot of the same distribution with outliers, minimum, 25th percentile, median, 75th percentile, and maximum.
To cite this abstract in AMA style:Brady K, Alexander R, Bloch R, Rudolph M, Baggiani K, Stimson D, Kammensheidt A. Estimation of the Prevalence of Hydroxychloroquine-Induced Retinopathy in a Cohort of Hydroxychloroquine-Compliant Patients [abstract]. Arthritis Rheumatol. 2021; 73 (suppl 9). https://acrabstracts.org/abstract/estimation-of-the-prevalence-of-hydroxychloroquine-induced-retinopathy-in-a-cohort-of-hydroxychloroquine-compliant-patients/. Accessed .
« Back to ACR Convergence 2021
ACR Meeting Abstracts - https://acrabstracts.org/abstract/estimation-of-the-prevalence-of-hydroxychloroquine-induced-retinopathy-in-a-cohort-of-hydroxychloroquine-compliant-patients/