Hydroxychloroquine Associated With Lower Glomerular Filtration Rate Decline in Lupus Nephritis

Abstract

Objective

Hydroxychloroquine (HCQ) protects kidney function in lupus nephritis (LN) by preventing flares, yet some cohort studies show no significant benefit in kidney function with HCQ. Clarifying these conflicting findings by showing early and long‐term benefits of HCQ on kidney function preservation is critical. Therefore, we analyzed data from our retrospective longitudinal inception LN cohort to examine the time‐varying effects of HCQ on kidney function decline in LN.

Methods

We analyzed retrospective data from an incident biopsy‐proven LN cohort. Creatinine values at LN diagnosis through the last follow‐up were abstracted to find the estimated glomerular filtration rate (eGFR). Using HCQ exposure as a time‐dependent covariate, we examined associations between HCQ exposure and sustained eGFR decline ≥30% and ≥40%. We also calculated an annual eGFR slope decline by HCQ exposure using linear mixed‐effects analysis.

Results

Among 209 patients, 33% and 23% experienced eGFR decline ≥30% and ≥40% over time. Time‐varying HCQ exposure was associated with a 60% and 62% lower risk of eGFR decline of ≥30% or ≥40%, after adjusting for propensity scores. A 77% lower risk of eGFR decline was noted in patients with chronic kidney disease (CKD) stage ≥3 with HCQ. HCQ exposure reduced the annual eGFR slope decline by 5.12 and 3.17 mL/min/1.73 m² within the first 5 and 10 years of diagnosis.

Conclusion

HCQ use was associated with early and long‐term benefits on kidney function in LN, including those with CKD stage ≥3. Universal HCQ use should be encouraged in LN patients.

INTRODUCTION

Lupus nephritis (LN) is the most common (50%–60%) severe life‐threatening manifestation of systemic lupus erythematosus (SLE). ¹ Up to 30% of patients with LN develop kidney failure, and kidney failure risk is 10‐ to 22‐fold higher (confidence interval [CI] 14–35) in patients with LN compared to age‐matched peers. ² Kidney failure often requires kidney replacement therapy, which negatively impacts the physical, social, financial, and emotional well‐being of patients with LN. ² , ³ , ⁴ Thus, preserving long‐term kidney function is a major therapeutic goal in LN. ⁵ Hydroxychloroquine (HCQ) is a foundational therapy in SLE treatment armamentarium, with several studies noting significant improvement in disease‐free and damage‐free survival in SLE with HCQ. Although guidelines for LN management recommend HCQ use unless contraindicated, the strength of this recommendation was historically modest (level C). ⁶ More recently, the 2024 Kidney Disease Improving Global Outcomes (KDIGO) and the American College of Rheumatology (ACR) LN guidelines strongly recommend HCQ use; however, the level of evidence supporting this recommendation remains modest, particularly for kidney outcomes. ⁷ , ⁸ Moreover, Wu et al reported a higher, but not statistically significant, risk of kidney function decline in HCQ users versus nonusers (hazard ratio [HR] 1.3, CI 0.4–4.3). ⁹ , ¹⁰ Such conflicting information could potentially explain low HCQ prescribing rates (66%–79%) and even lower prescribing rates by nonrheumatologists, particularly in patients with SLE limited to the kidneys (LN). ¹¹ , ¹²

SIGNIFICANCE & INNOVATIONS.

• This study highlights kidney‐specific effects of hydroxychloroquine (HCQ) exposure duration in lupus nephritis (LN) by showing a 60% and 62% lower risk of sustained estimated glomerular filtration rate (eGFR) decline of ≥30% and ≥40%, with HCQ exposure (status = yes) over time after adjusting for propensity scores.

• Particularly in patients with chronic kidney disease stage ≥3, a 77% lower risk of eGFR decline of ≥30% was noted with HCQ exposure.

• We are the first to show a yearly saving of 5.12 and 3.17 mL/min/1.73 m² in eGFR slope decline with HCQ within 5 and 10 years of LN diagnosis using linear mixed‐effects analysis with a random slope and a random intercept.

• Finally, these findings highlight both early and long‐term benefits of using HCQ in attenuating eGFR slope decline in LN and emphasize the importance of starting early and continuing HCQ therapy to preserve kidney health alongside overall lupus disease control.

A few observational studies and one small randomized controlled trial (including only pediatric LN) examined the impact of HCQ on kidney outcomes (eg, LN flares, kidney failure, doubling of serum creatinine). ¹¹ , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ However, pooled data analysis for LN guidelines noted imprecise risk estimates and uncertainty in the current evidence. ⁷ This may be due to a small number of events in existing studies given these studies used terminal events (eg, kidney failure or doubling of serum creatinine), which can take a long time to occur and could be missed due to losses to follow‐up or competing events. Thus, a recent consensus conference recommended using sustained kidney decline of ≥30% or ≥40% or annual estimated glomerular filtration rate (eGFR) slope decline as surrogate outcomes for eventual kidney failure to improve precision in risk estimates. ⁵ , ¹⁸ , ¹⁹ Moreover, data on the time‐varying impact of HCQ exposure on kidney function decline have not been examined thus far. ²⁰ Calculating an annual kidney function decline (eGFR slope decline) with and without HCQ could estimate the time‐varying treatment effect of HCQ on preserving kidney function over time in LN. This information will support clinicians and patients in their decisions to commit to prolonged HCQ therapy, even those with LN and limited extrarenal manifestations.

Therefore, using a retrospective biopsy‐proven inception LN cohort, we aimed to examine (1) associations between time‐varying HCQ exposure and sustained eGFR decline of ≥30% and ≥40% and (2) longitudinal time‐varying treatment effect of HCQ exposure on eGFR slope decline over time. Finally, we cross‐checked the longitudinal time‐varying treatment effect of HCQ on eGFR slope decline in a separate prevalent LN cohort identified using an electronic health records (EHR) data algorithm.

METHODS

Primary analysis using a retrospective biopsy‐proven inception LN cohort

Cohort Identification

In this retrospective cohort study, we identified all consecutive patients with lupus who underwent native kidney biopsy between 1994 and 2019 at the University of Wisconsin Hospital and Clinics (UW) and were diagnosed as having LN. We abstracted longitudinal visit‐level data since LN diagnosis until the last visit record from patients’ EHR and a comprehensive kidney biopsy database. We used the standard ACR SLE classification criterion ²¹ , ²² , ²³ and the 2003 International Society of Nephrology/Renal Pathology Society (ISN/RPS) LN classification to characterize patients with SLE and LN. ²⁴ The UW Human Research Protection Program approved this study with a waiver of informed consent (institutional review board numbers 2022‐0840 and 2016‐1260).

Variables

Sociodemographics & comorbidities. Using the EHR, we recorded sociodemographics at the time of biopsy (LN diagnosis) including age, sex, race, insurance type (commercial vs Medicaid or no insurance), and area deprivation index (ADI) derived from nine‐digit zip code (categorized as >80 vs ≤80; ADI >80 indicates neighborhoods with high socioeconomic status disadvantage). ²⁵ Data on comorbidities including hypertension (HTN), diabetes mellitus (DM), and hyperlipidemia, at or within the year of LN diagnosis, were abstracted, consistent with published literature. ²⁶ A standard validation method was used to ascertain comorbidities (HTN, DM, dyslipidemia), which included the following steps: (1) two or more encounters with International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, (ICD‐10) codes for the comorbidity documented ≥30 days apart AND (2) two abnormal test results for the comorbidity in two or more encounters OR (3) medication prescribed for the comorbidity.

LN biopsy findings. LN classes I through VI were abstracted from the initial kidney biopsy reports. ²⁴ LN class was categorized as proliferative, mixed (eg, class III ± V or IV ± V), and nonproliferative (eg, membranous, class I or II with or without podocytopathy). The NIH LN chronicity and activity indices were abstracted from the index LN biopsy reports. ²⁴ Renal arteriosclerosis was graded as percent luminal narrowing. ²⁷

LN disease course. We abstracted visit‐level data for laboratory measurements, such as serum creatinine and urine protein creatinine ratio (UPC), at LN diagnosis (at biopsy) until the last follow‐up visit records available in the EHR. We found the eGFR for each visit using the National Kidney Foundation recommended race‐neutral eGFR calculator (Chronic Kidney Disease Epidemiology Collaboration) and the abstracted serum creatinine values. Using KDIGO definitions for renal remission, we identified patients who achieved renal remission at 12 months after LN diagnosis. ⁷ Additionally, using standard definitions of LN flare, ⁷ we identified patients who experienced at least LN flare after diagnosis. SLE duration before LN diagnosis was categorized as less than two years or two or more years.

Time‐varying HCQ exposure. We used visit‐level medication prescription data from the EHR to ascertain HCQ use status (yes or no) at each time point. HCQ exposure status was ascertained as yes when there was an active HCQ prescription for at least three months at each follow‐up time point. Only the prescribed dose of HCQ was assessed in this paper; data on number of days covered by prescriptions were not available. Any change in HCQ exposure status at any given time point for each patient was accounted in analysis. ²⁸ To test the effect of a higher than guideline‐recommended HCQ dose, data on median HCQ dose over time were included as >5 versus ≤5 mg/kg/day categories.

Other medications. We also abstracted data on angiotensin converting enzyme inhibitor (ACE‐i) or angiotensin receptor blocker (ARB), aspirin, statin, and sodium–glucose cotransporter inhibitor ever use (yes/no). Next, we abstracted data on the type of induction and maintenance LN therapy regimen (cyclophosphamide vs mycophenolate vs combination therapies or others, such as belimumab, rituximab, calcineurin inhibitors, or azathioprine) and the duration of mycophenolate maintenance therapy (in years). Finally, glucocorticoid data were included as two categorical variables: (1) high‐dose glucocorticoids (60 mg or higher) at LN diagnosis (yes/no) and (2) prolonged glucocorticoid use defined as prednisone >7.5 mg (or equivalent doses of other steroids) for more than six months (yes/no). ²⁹

Outcomes. We analyzed three key outcomes.

1. Primary outcome: eGFR decline of ≥30%, defined as ≥30% reduction in eGFR for at least two follow‐up visits compared to baseline eGFR (at or within three months of LN diagnosis; a median value was calculated if multiple values were noted) or if a patient required sustained renal replacement therapy at any follow‐up time. ⁵ , ¹⁸ , ¹⁹

2. Secondary outcome: eGFR decline of ≥40%, defined as 40% or more reduction in eGFR for at least two follow‐up visits from baseline or if a patient required sustained renal replacement therapy at any follow‐up time. ⁵ , ¹⁸ , ¹⁹ The time to sustained eGFR decline of ≥30% or ≥40% was defined as time from LN diagnosis until the first follow‐up visit with ≥30% or ≥40% decline in eGFR from baseline.

3. Annual eGFR slope: visit‐level data on eGFR were used to estimate annual eGFR slope over follow‐up. The annual eGFR slope decline usually reflects the benefits of treatment and accounts for impact of treatment (HCQ) over time. All outcomes were abstracted by independent abstractors who were blinded to other variables (CS, LK, MK ).

Analysis

Descriptive data. The distribution of patient characteristics was examined. Based on distribution, continuous variables were expressed as mean ± SD, and categorical variables were expressed as n (%).

Data cleaning (missing data and censoring). Given that we abstracted visit‐level data from patients’ EHR, we ascertained loss of follow‐up if a patient had no visit‐level data, and these patients were censored from the analysis. Patients without sustained eGFR decline were censored at the end of the follow‐up. Visit‐level eGFR values, when used as an outcome, were not imputed. Standard methods, such as imputing median values for laboratory measurements, medications, and other variables, were used to handle missing data. ³⁰

Cox proportional hazards analysis using eGFR decline ≥30% as the primary outcome. We used a Cox proportional hazards (PH) model to analyze the time‐varying effects of HCQ use on eGFR decline of ≥30% (primary outcome). Both unadjusted and adjusted Cox regression analyses were conducted to examine associations between time‐dependent HCQ exposure and eGFR decline of ≥30%. ⁵ , ¹⁸ , ¹⁹ , ²⁸ The fully adjusted Cox model included a time‐dependent HCQ exposure variable and all relevant covariates that could influence eGFR decline. Schoenfeld residual analysis was performed, and variables such as insurance and history of HTN were stratified in the full Cox model to satisfy the PH assumption. Given the limited number of eGFR events, we also developed a reduced model that included key covariates: age, sex, race, insurance history of HTN (yes/no), baseline eGFR, NIH LN chronicity and activity indices, prolonged glucocorticoid use (yes/no), HCQ use status, and weight‐based HCQ dose categories (>5 mg/kg/day vs ≤5 mg/kg/day). Race and insurance were stratified to meet PH assumptions. E values for the effect size and 95% CIs were calculated. ³¹ The E value quantifies the minimum strength of association that an unmeasured confounder would need to have with both the exposure and the outcome for the observed association to be null. Higher E values (eg, >2) indicate that a strong unmeasured confounder would be needed to negate the observed effect, thereby supporting the robustness of the findings.

Cox PH analysis using eGFR decline ≥40% (secondary outcome). Similar methods as describe in the primary outcome section were used to examine the association between time‐dependent HCQ exposure and eGFR decline of ≥40%, adjusting for all covariables in the full model and only key variables in the limited model. Again, E values were calculated to account for unmeasured confounders.

Inverse probability weighting–adjusted models. To control for survival bias, we developed inverse probability weighting (IPW)‐adjusted models for both primary and secondary analyses. The propensity score was estimated using a logistic regression model with the following covariates: age, sex, race, insurance, history of HTN, baseline eGFR, NIH LN chronicity and activity indices, prolonged glucocorticoid use, mycophenolate mofetil (MMF) duration (in years), LN class, ACE‐i/ARB ever use, HCQ use at baseline (yes or no), and median HCQ dose categories. The weights were then used in the reduced Cox PH model to separately analyze the time to eGFR decline of ≥30% or ≥40%.

Sensitivity analysis. We performed the following sensitivity analyses: (1) excluding patients with follow‐up duration shorter than six months and (2) including all patients with at least 5 and 10 years of follow‐up. Doing so helped us examine early effects of HCQ use and have more uniformity in the cohort's follow‐up time. Additionally, variables such as LN remission and LN flares can have causal links with eGFR decline. Thus, these were not included in the primary and secondary analyses described earlier to avoid underestimation of treatment effect. However, these variables were added in separate models as part of the sensitivity analysis. Finally, we examined associations between time‐dependent HCQ exposure and eGFR decline of ≥30%, including only patients with eGFR ≤60 mL/min/1.73 m² (chronic kidney disease [CKD] stage ≥3) at LN diagnosis.

Linear mixed‐effects model to estimate annual eGFR slope decline. We used a linear mixed‐effects model with random intercepts and slopes (via the lme4 package in R [The R Project for Statistical Computing]) to analyze the time‐varying effect of HCQ on eGFR slope over time. The model included fixed effects for age, sex, race, history of HTN, NIH LN chronicity and activity indices, nephrotic syndrome at LN diagnosis (defined as UPC ≥3.5), CKD stage ≥3 at LN diagnosis, ACE‐i/ARB use (ever), time‐varying HCQ exposure, and time. Random effects were specified for both the intercept and the slope of time, allowing for individual variability in baseline eGFR and the rate of change over time. An interaction term between time and HCQ exposure was included to assess the time‐varying effect of treatment. To have more uniformity in the cohort's follow‐up time and examine early effect of HCQ exposure, follow‐up time for this model was limited to 10 years. To evaluate model fit and compare the inclusion of the time‐varying HCQ effect, we performed a likelihood ratio test between models with and without the interaction term. A visualization of eGFR trajectories for HCQ exposure (yes vs no) was generated using the ggplot2, cowplot, and effects packages in R.

Additional analysis in a prevalent cohort to test treatment effect of HCQ on EGFR slope

Prevalent LN cohort identification

Prevalent LN cases defined as those with two or more visits at least 30 days apart with an ICD‐10 code for LN (M32.14 or M32.15; 96% specificity) were identified in the EHR, and data were abstracted using an automated data query algorithm. ³²

Key variables

Using the definitions defined earlier, we abstracted data on sociodemographics (age, race, ethnicity, sex, insurance), comorbidities (history of HTN, DM, dyslipidemia), time‐varying HCQ exposure using prescription dates for HCQ, ACE‐i/ARB ever use, and laboratory values (creatinine to calculate eGFR and UPC to estimate nephrotic range proteinuria). Data for LN biopsy findings and LN disease course (induction regimen, maintenance regimen, and remission) were not available for prevalent cases.

Linear mixed‐effects analysis to cross‐check time‐varying treatment effect of HCQ on annual eGFR slope

We used a linear mixed‐effects model with random intercepts and slopes (as described earlier; lme4 in R) to analyze the time‐varying treatment effect of HCQ on eGFR slope over time, adjusting for sociodemographics (age, race, ethnicity, sex, insurance), ACE‐i/ARB, nephrotic range proteinuria, and history of HTN. ⁵ , ³³

RESULTS

Primary analysis using our retrospective inception LN cohort, N = 209

Retrospective inception LN cohort characteristics

Overall, 209 adults with biopsy‐proven incident LN were included in the cohort with a mean duration of follow‐up of 5.4 ± SD 3.8 years (Table 1). A total of 66 and 48 patients experienced sustained eGFR decline ≥30% or ≥40% over time; 35 patients experienced eGFR decline of ≥30% by year five. Key patient characteristics are highlighted in Table 1.

Table 1.

Baseline patient characteristics (N = 209)^*

Variables	Value
Age at diagnosis, mean ± SD, y	38 ± 15
Sex, n (%)
Female	156 (75)
Male	53 (25)
Race and ethnicity, n (%)
Asian	30 (14)
Black	24 (12)
Hispanic	15 (7)
White	140 (67)
Medicaid or no insurance, n (%)	72 (35)
ADI ≥ 80, n (%)	27 (13)
eGFR at LN diagnosis, mean ± SD, mL/min/1.73 m2	80 ± 37
SLE duration ≥2 y before LN diagnosis, n (%)	59 (28)
HTN at/within 1 y of diagnosis, n (%)	82 (39)
Diabetes at/within 1 y of diagnosis, n (%)	11 (5)
UPC at LN diagnosis, mean ± SD, mg/mg	2.9 ± 3.3
LN class per index LN biopsy, n (%)
Proliferative	108 (52)
Nonproliferative	55 (26)
Mixed (eg, LN class IV/V)	46 (22)
LN activity index per index LN biopsy, mean ± SD	2.5 ± 3.3
LN chronicity index per index LN biopsy, mean ± SD	2.2 ± 2.1
Renal arteriosclerosis in % luminal area narrowing, mean ± SD	22 ± 28
Renal remission at 12 mo, n (%)
Complete remission	144 (69)
Partial remission	33 (16)
High dose of steroids at diagnosis, n (%)	112 (50)
Induction therapy after diagnosis, n (%)
Cyclophosphamide	44 (21)
Mycophenolate	146 (70)
Combination or others	19 (9)
Maintenance therapy after diagnosis, n (%)
Mycophenolate	155 (74)
Azathioprine or others	54 (26)
Mycophenolate use duration, mean ± SD, y	3.7 ± 3.2
Chronic steroid (≥7.5 mg/d for ≥6 mo), n (%)	88 (42)
HCQ exposure status = yes at diagnosis, n (%)	166 (79)
Higher than guideline‐recommended HCQ dose >5 mg/kg/d, yes, n (%)	39 (19)
ACE‐i/ARB use, n (%)	140 (67)
SGLT2i use, n (%)	13 (6)
Follow‐up time, mean ± SD, y	5.4 ± 3.8
Follow‐up time >5 y, n (%)	108 (44)

^*eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration formular. ACE‐i, angiotensin converting enzyme inhibitor; ADI, area deprivation index; ARB, angiotensin receptor blocker; eGFR, estimated glomerular filtration rate; HCQ, hydroxychloroquine; HTN, hypertension; LN, lupus nephritis; SGLT2i, sodium–glucose cotransporter 2 inhibitor; SLE, systemic lupus erythematosus; UPC, urine protein creatinine.

Time‐varying HCQ exposure and eGFR decline of ≥30% (primary outcome)

Using unadjusted univariable Cox regression models and HCQ exposure as a time‐dependent covariate, HCQ exposure (status = yes) was associated with a 50% lower rate of sustained eGFR decline of ≥30% (unadjusted HR 0.50 [95% CI 0.29–0.86]; P = 0.012; Supplementary Table 1). After adjusting for covariables, time‐varying HCQ exposure status (yes) was associated with 59% (adjusted HR 0.41 [95% CI 0.22–0.78]; Table 2A) and 64% (adjusted HR 0.36 [95% CI 0.19–0.68]; Table 2B) lower rates of achieving the primary outcome (eGFR decline of ≥30%) in the full and reduced models, respectively. Given sample size limitations and no association between LN specific treatments and eGFR decline in unadjusted models (Supplementary Table 1), LN specific treatments were not included in the adjusted models. Additionally, an E value of 5.0 was calculated for the reduced model, indicating that any unmeasured confounder needs to be associated with five‐fold higher eGFR decline of ≥30% for the association with HCQ exposure to be truly null in the reduced model. Even after adjusting for propensity score, the risk of eGFR decline of ≥30% significantly decreased with HCQ exposure (IPW adjusted HR 0.40 [95% CI 0.23–0.69]; P = 0.001; Table 3A).

Table 2.

Adjusted Cox regression model showing associations between time‐varying HCQ exposure and eGFR decline of ≥30% (N = 209)^*

Variables	Adjusted HR (95% CI) a	P value
Full multivariable Cox model a
Age in 1‐y increments	1 (0.98–1.02)	0.70
Female	0.81 (0.43–1.52)	0.51
White race	ref
Asian race	1.88 (0.80–4.43)	0.15
Black race	4.20 (1.86–9.48)	0.0005
Hispanic ethnicity	0.76 (0.17–3.32)	0.72
eGFR per 10 mL/min/1.73 m2 decrease	1.09 (0.98–1.20)	0.11
NIH LN chronicity index per 1‐point increase	1.28 (1.10–1.50)	0.002
NIH LN activity index per 1‐point increase	1.12 (1.04–1.22)	0.005
Renal arteriosclerosis per 1% increase	1.01 (0.99–1.02)	0.39
MMF exposure duration per 1‐y increase	0.94 (0.86–1.03)	0.20
ACE‐i/ARB ever use, yes	0.87 (0.49–1.53)	0.62
Prolonged glucocorticoids use (>7.5 mg >6 mo), yes c	1.74 (0.96–3.16)	0.07 d
HCQ exposure (status = yes) e	0.41 (0.22–0.78)	0.007
Higher than guideline recommended HCQ dose >5 mg/kg/d, yes f	1.08 (0.55–2.23)	0.83
Reduced multivariable Cox model b
Age in 1‐y increments	1.01 (0.99–1.03)	0.40
Female	0.87 (0.47–1.62)	0.67
History of HTN, yes	1.72 (0.98–3.04)	0.06 d
eGFR per 10 mL/min/1.73 m2 decrease	1.09 (0.98–1.2)	0.10
NIH LN chronicity index per 1‐point increase	1.34 (1.16–1.54)	0.0001
NIH LN activity index per 1‐point increase	1.13 (1.03–1.23)	0.007
Prolonged glucocorticoids use, yes c	1.86 (1.01–3.44)	0.047
HCQ exposure (status = yes) e	0.36 (0.19–0.68)	0.002
Higher than guideline‐recommended HCQ dose >5 mg/kg/d, yes f	1.46 (0.71–3.03)	0.30

^*eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration formular. Variables with a significant P value (<0.05) are shown in bold font. ACE‐i, angiotensin converting enzyme inhibitor; ARB, angiotensin receptor blocker; CI, confidence interval; eGFR, estimated glomerular filtration rate; HCQ, hydroxychloroquine; HR, hazard ratio; HTN, hypertension; LN, lupus nephritis; MMF, mycophenolate mofetil; PH, proportional hazards.

^a Adjusted for age, race, sex, insurance type/status at diagnosis, history of HTN, eGFR at LN diagnosis, NIH LN chronicity and activity indices, mycophenolate duration, ACE‐i use, prolonged glucocorticoids use (defined as >7.5 mg of prednisone or equivalent doses for >6 mo), HCQ exposure status, and weight‐based HCQ dose categories; insurance and history of HTN were stratified based on Schoenfeld residual analysis to satisfy PH assumption for this model (global PH > 0.05).

^b Adjusted for age, race, insurance type/status at diagnosis, sex, history of HTN, eGFR at LN diagnosis, NIH LN chronicity and activity indices, prolonged glucocorticoids use, HCQ exposure status, and weight‐based HCQ dose categories; race and insurance status/type were stratified based on Schoenfeld residual analysis to satisfy PH assumption (global PH > 0.05).

^c Defined as glucocorticoids use ≥7.5 mg for >6 mo after LN diagnosis.

^d Trend toward statistical significance.

^e HCQ exposure status was abstracted and used as a time‐dependent covariate.

^f HCQ dose >5 mg/kg/d is compared to a combination of HCQ dose ≤5 mg/kg/d or no HCQ use.

Table 3.

IPW‐adjusted and crude Cox models showing associations between HCQ exposure and HCQ dose with sustained eGFR decline of ≥30% and ≥40%, N = 209^*

	Adjusted HR (95% CI)	P value
Crude vs IPW‐adjusted Cox models for eGFR decline ≥30%
Crude adjusted Cox model a
HCQ exposure (status = yes)	0.36 (0.19–0.68)	0.0016
Higher than guideline‐recommended HCQ dose >5 mg/kg/d, yes e	1.46 (0.71–3.03)	0.30
IPW‐adjusted b model
HCQ exposure (status = yes)	0.40 (0.23–0.69)	0.001
Higher than guideline‐recommended HCQ dose >5 mg/kg/d, yes e	2.02 (1.13–3.62)	0.02
Crude vs IPW‐adjusted Cox models for eGFR decline ≥40%
Crude adjusted Cox model c
HCQ exposure (status = yes)	0.33 (0.17–0.65)	0.002
Higher than guideline‐recommended HCQ dose >5 mg/kg/d, yes e	1.18 (0.53–2.66)	0.68
IPW‐adjusted d model
HCQ exposure (status = yes)	0.38 (0.21–0.70)	0.002
Higher than guideline‐recommended HCQ dose >5 mg/kg/d, yes e	1.49 (0.75–2.96)	0.26

^*Variables with a significant P value (<0.05) are shown in bold font. ACE‐i, angiotensin converting enzyme inhibitor; ARB, angiotensin receptor blocker; eGFR, estimated glomerular filtration rate; HCQ, hydroxychloroquine; HR, hazard ratio; HTN, hypertension; IPW, inverse probability weighting; LN, lupus nephritis; MMF, mycophenolate mofetil; PH, proportional hazards.

^a Adjusted for age, race, insurance type/status at diagnosis, sex, history of HTN, eGFR at LN diagnosis, NIH LN chronicity and activity indices, prolonged glucocorticoids use, HCQ exposure status, weight‐based HCQ dose categories; race and insurance status/type were stratified based on Schoenfeld residual analysis to satisfy PH assumption (global PH > 0.05).

^b IPW models adjust for propensity scores for the following covariates: age, sex, race, insurance, history of HTN, baseline eGFR, NIH LN chronicity and activity indices, prolonged glucocorticoid use, MMF duration (in years), LN class, ACE‐i/ARB ever use, baseline HCQ use, and median weight‐based HCQ dose categories.

^c Adjusted for age, race, insurance type/status at diagnosis, sex, history of HTN, eGFR at LN diagnosis, NIH LN chronicity and activity indices, HCQ exposure status, and weight‐based HCQ dose categories.

^d IPW models adjust for propensity scores for the following covariates: age, sex, race, insurance, history of HTN, baseline eGFR, NIH LN chronicity and activity indices, prolonged glucocorticoid use, MMF duration (in years), LN class, ACE‐i/ARB ever use, baseline HCQ use, and median weight‐based HCQ dose categories.

^e HCQ dose >5 mg/kg/d is compared to a combination of HCQ dose ≤5 mg/kg/d or no HCQ use.

HCQ exposure and eGFR decline of ≥40% (secondary outcome)

In an unadjusted Cox model, time‐varying HCQ exposure was associated with a 58% lower likelihood of sustained eGFR decline of ≥40% (unadjusted HR 0.42 [95% CI 0.23–0.77]; P = 0.005; Supplementary Table 2). After adjusting for known covariables, HCQ exposure remained strongly associated with a 66% and 67% lower risk of eGFR decline of ≥40% in both full and reduced Cox models (Table 4). An E value of 5.5 was calculated for the reduced model. Finally, a strong association between HCQ exposure and eGFR decline of ≥40% remained even after adjusting for propensity scores (IPW‐adjusted HR 0.38 [95% CI 0.21–0.70]; P = 0.002; Table 3B).

Table 4.

Adjusted Cox regression model showing associations between time‐varying HCQ exposure and eGFR decline of ≥40% (N = 209)^*

Variables	Adjusted HR (95% CI) a	P value
Full multivariable Cox model a
Age in 1‐y increments	1.01 (0.99–1.03)	0.57
Female	1.2 (0.56–2.59)	0.64
History of HTN, yes	2.12 (1.09–4.15)	0.027
eGFR per 10 mL/min/1.73 m2 decrease	1.07 (0.95–1.20)	0.27
NIH LN chronicity index per 1‐point increase	1.33 (1.13–1.57)	0.001
NIH LN activity index per 1‐point increase	1.10 (1.00–1.22)	0.053 b
MMF exposure duration per 1‐y increase	0.95 (0.85–1.07)	0.40
ACE‐i/ARB ever use, yes	0.77 (0.40–1.51)	0.45
Prolonged glucocorticoids use, yes c	1.66 (0.84–3.32)	0.15
HCQ exposure (status = yes) d	0.34 (0.16–0.70)	0.004
Higher than guideline‐recommended HCQ dose >5 mg/kg/d, yes e	1.55 (0.66–3.65)	0.32
Reduced multivariable Cox model f
Age in 1‐y increments	1.00 (0.98–1.03)	0.66
Female	1.21 (0.60–2.41)	0.60
White race	ref
Asian race	1.45 (0.59–3.57)	0.75
Black race	1.20 (0.40–3.60)	0.42
Hispanic ethnicity	1.09 (0.25–4.83)	0.91
No insurance or Medicaid	2.68 (1.42–5.06)	0.002
History of HTN, yes	1.76 (0.94–3.28)	0.08 d
eGFR per 10 mL/min/1.73 m2 decrease	1.07 (0.96–1.20)	0.23
NIH LN chronicity index per 1‐point increase	1.30 (1.12–1.52)	0.001
NIH LN activity index per 1‐point increase	1.12 (1.01–1.23)	0.03
HCQ exposure (status = yes) d	0.33 (0.17–0.65)	0.002
Higher than guideline‐recommended HCQ dose >5 mg/kg/d, yes e	1.18 (0.53–2.66)	0.68

^*eGFR calculated using the Chronic Kidney Disease Epidemiology Collaboration formular. Variables with a significant P value (<0.05) are shown in bold font. ACE‐i, angiotensin converting enzyme inhibitor; ARB, angiotensin receptor blocker; CI, confidence interval; eGFR, estimated glomerular filtration rate; HCQ, hydroxychloroquine; HR, hazard ratio; HTN, hypertension; LN, lupus nephritis; MMF, mycophenolate mofetil; PH, proportional hazards.

^a Adjusted for age, race, sex, insurance type/status at diagnosis, history of HTN, eGFR at LN diagnosis, NIH LN chronicity and activity indices, mycophenolate duration, ACE‐i use, prolonged glucocorticoids use (defined as >7.5 mg of prednisone or equivalent doses for >6 mo), HCQ exposure status, and weight‐based HCQ dose categories; race and insurance were stratified based on Schoenfeld residual analysis to satisfy PH assumption for this model (global PH > 0.05).

^b Indicates trend toward significance.

^c Defined as glucocorticoids use ≥7.5 mg for >6 mo after LN diagnosis.

^d HCQ exposure status was abstracted and used as a time‐dependent covariate.

^e HCQ dose >5 mg/kg/d is compared to a combination of HCQ dose ≤5 mg/kg/d or no HCQ use

^f Adjusted for age, race, insurance type/status at diagnosis, sex, history of HTN, eGFR at LN diagnosis, NIH LN chronicity and activity indices, HCQ exposure status, weight‐based HCQ dose categories, and PH assumption (global PH > 0.05).

Sensitivity analysis

All patients had at least six months of follow‐up. No significant changes were noted in associations between time‐varying HCQ exposure with eGFR decline of ≥30% after censoring time to 5 and 10 years after LN diagnosis (Supplementary Table 3A–B).

Even after adjusting for LN remission and flares, time‐varying HCQ exposure was associated with a 63% lower risk of eGFR decline of ≥30% (adjusted HR 0.37 [95% CI 0.31–0.47]; P = 0.005; not shown in a table). Finally, in a subgroup analysis including only patients with CKD stage ≥3 at LN diagnosis, HCQ exposure was associated with a 77% lower risk of eGFR decline of ≥30% even in this high‐risk group (adjusted HR 0.23 [95% CI 0.06–0.83]; P = 0.03; Supplementary Table 4).

Time‐varying treatment effect of HCQ on annual eGFR slope decline

Average number of observations per patient were similar in groups with and without HCQ exposure (Figure 1). Adjusted annual eGFR slope decline with time‐varying HCQ exposure was −0.65 mL/min/1.73 m² (95% CI −0.06 to −1.26 mL/min/1.73 m²) over 10 years (Figure 1A), whereas the adjusted annual eGFR slope decline without HCQ was five times steeper (−3.82 mL/min/1.73 m²/year [95% CI −1.20 to −6.44 mL/min/1.73 m²/year]; Figure 1B) in the first 10 years after diagnosis. A significant reduction in annual eGFR slope decline by 3.17 mL/min/1.73 m²/year (P < 0.0001) was noted with HCQ exposure. A yearly saving of 5.12 mL/min/1.73 m² in the adjusted annual eGFR slope decline was noted within 5 years of diagnosis with HCQ (Supplementary Figure 1).

Figure 1.

(A) Annual eGFR slope decline with HCQ use versus (B) without HCQ use, using random slope and intercept linear mixed‐methods model adjusting for covariables (age, race, sex, LN pathology findings, comorbidities, treatment) in our LN inception cohort (N = 209). CI, confidence interval; eGFR, estimated glomerular filtration rate; HCQ, hydroxychloroquine; LN, lupus nephritis.

Testing treatment effect of HCQ on eGFR slope using a prevalent LN cohort

Prevalent LN cohort characteristics

Key characteristics of the prevalent LN cohort (N = 819) were as follows: mean age 54 ± SD 17 years, 59% women, 83% of White race, 11% of Black race, 4% of Hispanic ethnicity, and 6% having Medicaid. At the first visit, mean eGFR was 67 ± SD 31 mL/min/1.73 m², 24% had HTN, 4% had nephrotic syndrome, 71% were ACE‐i users, and 32% had been exposed to HCQ.

Treatment effect of HCQ on annual eGFR slope decline in the prevalent cohort

Similar time‐varying effects were noted with HCQ exposure on annual eGFR slope decline in a separate prevalent LN cohort with longer follow‐up. The adjusted annual eGFR slope decline with HCQ was half the annual slope decline noted without HCQ exposure (−0.94 vs 2.23 mL/min/1.73 m²/year; P < 0.0001; Figure 2A–B). Time‐varying treatment effect of HCQ on adjusted annual eGFR slope decline was 1.26 mL/min/1.73 m² (Figure 2).

Figure 2.

Linear mixed‐effects model examining the treatment effect of HCQ on eGFR in a prevalent LN cohort (N = 819 patients, including all visits of patients with ≥2 ICD‐10 LN diagnosis codes controlling for confounders [sociodemographics, comorbidities, LN characteristics, ACE‐i exposure]). All data points shown in both graphs. CI, confidence interval; eGFR, estimated glomerular filtration rate; HCQ, hydroxychloroquine; ICD‐10, International Statistical Classification of Diseases and Related Health Problems, Tenth Revision; LN, lupus nephritis. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/acr.25616/abstract.

DISCUSSION

Kidney involvement is one of the most common severe manifestations of SLE that occurs in up to 70% of patients, and up to 30% of patients experience irreversible kidney damage requiring renal replacement therapy. Literature in SLE highlights lower LN flare risk and prolonged survival with HCQ use, yet information regarding the time‐varying treatment effect of HCQ use on preservation of kidney function (eGFR) is limited. Our study leveraged a retrospective biopsy‐proven incident LN cohort and reported 59% and 66% lower risk of sustained eGFR decline of ≥30% and ≥40% with time‐varying HCQ exposure. Moreover, HCQ exposure was associated with 77% lower risk of eGFR decline of ≥30% in patients with CKD stage ≥3. Using linear mixed‐effects modeling to estimate the treatment effect of HCQ on annual eGFR slope, we are the first to report a yearly saving in annual eGFR slope decline by 5.12 and 3.17 mL/min/1.73 m² with HCQ within 5 and 10 years of diagnosis using our incident LN cohort. A similar reduction in annual eGFR slope decline by 1.26 mL/min/1.73 m² was noted with HCQ exposure in a separate large EHR‐based prevalent LN cohort with longer follow‐up. These findings directly address existing gaps in the literature, reduce uncertainty regarding long‐term impact of HCQ use on kidney function, and highlight a benefit of early initiation of HCQ in LN, thereby empowering clinicians to advocate for universal use HCQ in LN.

There are limited data on the longitudinal impact of HCQ on kidney function decline over time. ⁹ , ²⁰ Moreover, traditional clinical end points, such as kidney failure or doubling of serum creatinine, require large sample sizes and long follow‐up, resulting in insufficient power and low precision in current literature. ⁵ , ⁷ , ⁹ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ Using surrogate end points such as eGFR slope overcomes these limitations, and eGFR slope analysis can examine the impact of treatment on preservation of kidney function over time. Our study directly addresses this gap by leveraging a linear mixed‐effects model to estimate the time‐varying treatment effect of HCQ exposure. This analysis revealed that HCQ exposure predicted an attenuation in adjusted annual eGFR slope decline by 3.17 mL/min/1.73 m²/year (P < 0.0001) over 10 years after LN diagnosis in our incident biopsy‐proven LN cohort. More importantly, a yearly saving of 5.12 mL/min/1.73 m² was noted in eGFR slope decline with HCQ exposure within the first five years of LN diagnosis, indicating early treatment benefits. Finally, using a separate prevalent LN cohort with longer follow‐up, HCQ use again was associated with an attenuation in annual eGFR slope by 1.26 mL/min/1.73 m² over a longer follow‐up period. A meta‐regression found that an attenuation in the rate of decline in eGFR of ≥0.75 mL/min/1.73 m² per year can reduce the risk of kidney failure progression by 16% to 23%. ⁵ , ³⁸ Thus, our observed attenuation in eGFR decline over different time periods (range = 1.26–5.12 mL/min/1.73 m²) would be associated with at least 23%, if not more, reduced kidney failure risk. Moreover as an adjunct therapy for LN, HCQ could enhance effectiveness of LN induction and maintenance therapies.⁵ Thus, our study uniquely highlights the critical impact of HCQ use on preserving kidney function and underscores the need to use HCQ early and for a longer duration in LN.

Existing literature shows an overall positive association between HCQ use and preventing renal damage. ¹¹ , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ Particularly, Pons‐Estel et al noted a 70% reduction in the risk of renal damage (defined as 50% reduction in eGFR) with HCQ use. ³⁴ Likewise, recent studies using cohorts from Israel and Mexico, reported 0.4‐fold lower risk of CKD progression to stage ≥3 or kidney failure in HCQ users. ¹⁴ , ¹⁶ Despite strong associations and universal recommendation to use HCQ in LN, <70% of patients get started on HCQ. ¹¹ , ¹² , ⁴⁰ Underuse of HCQ could be linked to current gaps in literature: (1) low precision in risk estimates (wide CIs) given few terminal events (kidney failure or doubling of serum creatinine or CKD progression) occurred over time, ⁷ (2) contrasting evidence when time was used as a covariable highlighting no association or worsening of podocyte function with HCQ in LN possibly due to immortal time and survival biases, ¹⁰ , ⁴¹ (3) timing and duration of HCQ therapy is not clear, and (4) treatment effect of HCQ in CKD stage ≥3 needs clarification. ⁹ , ⁴⁰ Such gaps can exacerbate existing fears to use HCQ, particularly in nonrheumatology clinicians. ⁴⁰ Our study directly addresses these four gaps and bolsters the evidence regarding HCQ's specific benefit in LN. First, by using sustained eGFR ≥30% or ≥40% and eGFR slope decline as surrogate end points of kidney failure, we were able to improve precision of risk estimates. Second, even after propensity score adjustments and using HCQ exposure as a time‐dependent covariate, a significant 54% and 51% reduction in end points, eGFR decline of ≥30% and ≥40%, was noted with HCQ exposure. Third, a significant attenuation in eGFR slope and a 73% lower risk of eGFR decline of ≥30% was noted with early HCQ use, within five years of diagnosis. Finally, our subgroup analysis highlighted a 77% lower risk of eGFR decline of ≥30% with HCQ use in patients with CKD stage ≥3. By addressing the concerns about why, when, and how long HCQ should be used in LN, including those with CKD, our findings can support clinicians and patients in shared decision‐making, encouraging higher uptake of HCQ.

This study has several strengths including using the following: (1) a validated incident biopsy‐proven LN cohort, (2) HCQ exposure as a time‐dependent covariate, and (3) a robust linear mixed‐effects analysis to determine treatment effect of HCQ on eGFR slope decline. However, we acknowledge limitations. First, this study uses data from a single academic institution and may not represent the LN population in the US. Second, we did not have data on HCQ blood levels to account for variability in adherence and absorption of HCQ. Third, we could have missed eGFR decline in patients who moved, lost follow‐up, or died in outside institutions; however, several sensitivity analyses were performed. Fourth, given the retrospective design of the study, we could not account for unknown confounders and establish causal associations between HCQ exposure and kidney function decline. However, we calculated E values of 5 and 6, indicating that any unmeasured confounder needs to be associated with five‐fold and six‐fold higher eGFR decline of ≥30% and ≥40% for the association with HCQ exposure duration to be truly null. Additionally, we performed IPW analysis and Cox analysis using HCQ exposure as a time‐dependent covariate to address survival and immortal time bias. Fifth, our CKD stage ≥3 subgroup was small, and our findings need to be validated in larger cohorts. Finally, we could not ascertain the reasons for stopping HCQ or delayed onset of HCQ due to limitations of the study design.

The implications of our findings are three‐fold. First, they highlight kidney‐specific effects of HCQ exposure duration in LN, particularly on eGFR decline ≥30% or ≥40%, and reduction in annual eGFR slope decline (early surrogate end points of terminal events like kidney failure). Second, our study supports the inclusion of HCQ as a pivotal component of LN management protocols, not just for its systemic benefits, but also for a potential to preserve kidney function. Finally, these findings highlight both early and long‐term benefits of using HCQ in attenuating eGFR slope decline in LN. These findings emphasize the importance of starting early and continuing HCQ therapy to preserve kidney health alongside overall disease control even if SLE is limited to the kidneys (LN).

AUTHOR CONTRIBUTIONS

All authors contributed to at least one of the following manuscript preparation roles: conceptualization AND/OR methodology, software, investigation, formal analysis, data curation, visualization, and validation AND drafting or reviewing/editing the final draft. As corresponding author, Dr Garg confirms that all authors have provided the final approval of the version to be published, and takes responsibility for the affirmations regarding article submission (eg, not under consideration by another journal), the integrity of the data presented, and the statements regarding compliance with institutional review board/Declaration of Helsinki requirements.

Supporting information

Disclosure form.

Appendix S1: Supplementary Information