Susan Manzi

This study attempted to quantify the bias expected due to partly interval censored (IC) outcomes in the estimated association between hydroxychloroquine (HCQ) taper/cessation and time to disease flare among individuals with systemic lupus erythematosus (SLE). Using data-driven simulations, we estimated bias expected due to IC using real-world data from the Systemic Lupus International Collaborating Clinics inception cohort. The time-varying exposure of interest was a binary indicator of hydroxychloroquine tapering/cessation. The composite outcome was lupus flare, defined as lupus hospitalizations or increases in disease activity or medication dose. The two latter components were IC, as they were recorded only at annual assessment, without a precise date. For the unknown IC event times, a "true" event time was randomly generated from a uniform distribution of the time between two assessments. Each simulated sample was analyzed separately imputing unknown event times (for IC outcomes) either at the midpoint or endpoint of the interval between the two adjacent yearly assessments. Results of multivariable Cox proportional hazards models, adjusted for demographics, drugs, and clinical variables, using either "true" or imputed IC event times were compared. The 1,543 SLE patients were followed for a median of 42.2 months. During follow-up, 396 participants tapered/stopped HCQ and 1,187 experienced a flare. The adjusted uncorrected hazard ratio (HR) was 1.51 (95%CI: 1.30, 1.75) and 1.40 (95%CI: 1.21, 1.62) for midpoint and endpoint imputations, respectively. Data-driven simulations showed that imputation of IC event times resulted in a small but systematic bias toward the null that was consistently larger for endpoint than for midpoint imputation. IC events induced bias toward the null in the estimated association between HCQ taper/cessation and lupus flares. Data-driven simulations are useful for quantitative bias analyses in complex situations, as they allow accounting for relevant characteristics of a particular real-world dataset.

The multinational, phase 3, double-blind, placebo-controlled TULIP-SC trial evaluated the efficacy and safety of subcutaneous anifrolumab in adults who have moderate-to-severe SLE activity, despite receiving standard therapy. Adults with SLE received subcutaneous anifrolumab 120 mg or placebo once weekly for 52 weeks (1:1 randomization). Only the primary endpoint (treatment difference in BILAG-based Composite Lupus Assessment [BICLA] response at 52 weeks) was formally tested in a pre-planned interim analysis; key secondary and other endpoints were tested in the full analysis. At the interim analysis (220 patients, anifrolumab: n=109; placebo: n=111), the primary endpoint was met (anifrolumab vs placebo: 59.4% vs 43.9%; BICLA response difference [95% confidence interval]=15.5% [2.3-28.6%], p=0.0211). The full analysis included 367 patients (anifrolumab: n=184; placebo: n=183). Versus placebo, more patients treated with anifrolumab attained a BICLA response whilst maintaining low/reduced oral glucocorticoid doses through Week 52 (56.2% vs 34.0%; difference=22.3% [12.3-32.2] p<0.0001), and the time to first sustained BICLA response was reduced (Hazard ratio=2.2 [1.5-3.2]; p<0.0001). Treatment differences in Week 52 DORIS remission and Low Lupus Disease Activity State attainment rates favored anifrolumab over placebo (14.2% [5.6-22.8%], p=0.0012, and 14.1% [4.6-23.6%], p=0.0038). The frequencies of serious adverse events were 11.9% with anifrolumab and 10.4% with placebo; the frequencies of herpes zoster were 3.8% and 1.1%, respectively. Consistent with the well-established profile of intravenous anifrolumab, subcutaneous anifrolumab demonstrated significant, clinically meaningful treatment benefits when added to standard therapy, and an acceptable safety profile in patients with moderate to severe SLE.

Systemic lupus erythematosus (SLE) remains a deadly disease, yet our ability to predict adverse outcomes is poor. Mitochondria are organelles recognised by the immune system when released from cells, and antimitochondrial antibodies (AMA) can be detected in people with SLE. We assessed AMA as markers of nephritis, arterial vascular events (AVE), and other outcomes including mortality. We studied sera and data from 1114 participants of the Systemic Lupus International Collaborating Clinics inception cohort. We measured antiwhole mitochondria (AwMA), antimitochondrial DNA (AmtDNA), and antimitochondrial RNA (AmtRNA) antibodies by direct Enzyme-Linked ImmunoSorbent Assays (ELISAs). Separate multivariable Cox proportional hazards regression models estimated associations of either baseline or most recent measures of AMA with the outcomes, adjusted for biological sex, age, medications, and other clinical factors. Interactions of AMA with biological sex were tested for each outcome. All AMA titres were elevated in SLE vs healthy individuals. Higher AMA levels were associated with a higher hazard of nephritis, with the strongest associations for most recent AmtDNA (adjusted hazard ratio [aHR] =1.61 for increase of 1 SD, 95% CI 1.43-1.82) and AmtRNA (aHR =1.59, 1.46-1.73). Higher baseline AwMA levels predicted early mortality (aHR =1.19, 1.02-1.40). Most recent AmtDNA (aHR =1.68, 1.28-2.19) was associated with higher mortality throughout the follow-up. For AVE, the impact of higher AmtRNA was stronger in females. Baseline and most recent assessments of AMA levels may help identify individuals at higher risk of severe outcomes in SLE, including mortality. Integrating AMA into precision medicine strategies will allow deeper exploration of lupus heterogeneity.

Using hydroxychloroquine (HCQ) dose of 5 mg/kg/day in systemic lupus erythematosus (SLE) is associated with a higher risk of flares; HCQ blood level monitoring could be a better way to adjust HCQ dose. We studied the upper threshold for a reference range of HCQ levels to inform routine monitoring. This observational study included patients (n=2010) across the Systemic Lupus International Collaborating Clinics (SLICC), Wisconsin, International, and French studies, who underwent HCQ blood level measurements. Using adjusted spline and logistic regression analyses on the cross-sectional data, we first identified a HCQ blood level associated with higher HCQ toxicity. Next, we tested if this upper threshold level was supratherapeutic (no further risk reduction for SLE Disease Activity Index 2000 (SLEDAI-2K ≥6). Finally, we examined associations between chronic kidney disease (CKD) stage and supratherapeutic (toxic) HCQ blood levels. Among 1842 patients (excluding 168 patients with very low HCQ blood levels), 4.9% had HCQ related toxicity. Odds of toxicity were 2.1-fold higher with blood levels ≥1150 ng/mL, and 1.7-fold higher with cumulative HCQ dose per 1000g increase. Blood levels ≥1150 ng/mL were associated with a saturation in therapeutic effect, indicating supratherapeutic levels. Patients with CKD stage ≥3 had 2.3-fold higher odds of having supratherapeutic levels (≥1150 ng/mL). The therapeutic reference range for HCQ blood level monitoring is 750-<1150 ng/ml. HCQ level monitoring could optimize HCQ use, particularly in patients with CKD stage ≥3. Future longitudinal studies are needed to validate the use of HCQ blood level monitoring in optimizing dosing.

We aimed to compare clinical outcomes between patients in the Allegheny Health Network rheumatoid arthritis (RA) care pathway and patients receiving usual care.
The care pathway initiative implements guideline-based best practice alongside multi-disciplinary team-based care. Clinical and insurance claims data were extracted to compare the proportion of patients in clinical disease activity index (CDAI)-based remission and evaluate health care utilization and per member per month (PMPM) costs of care.
The RA care pathway cohort included 817 patients, and the usual care cohort included 3651 patients. The care pathway cohort had a significantly higher proportion of patients achieving remission at 6, 12, and 24 months (33.5% vs 20.3%, hazard ratio 1.53, 95% confidence interval 1.11-2.10, p = 0.008 at 24 months), and on average, 64 days earlier than usual care. This trend was also observed when including low disease activity patients with patients in remission and in a subgroup analysis of newly-diagnosed RA. RA-related PMPM costs were significantly higher in the care pathway group and primarily related to higher baseline CDAI, comorbidities, and biologic/targeted synthetic disease modifying anti-rheumatic drug use and switching.
Patients in our RA care pathway were more likely to achieve remission than usual care. PMPM RA costs were higher compared to usual care. We plan to use a longer follow-up to investigate if improved clinical outcomes result in reduced direct and indirect costs, to study the impact of individual team-based care interventions, and to validate our findings in other populations.

Background/Purpose Antinuclear antibody (ANA) testing is used to screen for systemic autoimmune rheumatic diseases (SARD) like systemic lupus erythematosus. It is well established that a nuclear dense fine-speckled (DFS) ANA pattern (AC-2), being rare among SARD patients, decreases the likelihood of these conditions. However, the AC-2 pattern is challenging for lab technologists to accurately identify due to similarities with other patterns, ie, AC-4 (speckled) and AC-30 (nuclear speckled with mitotic plate staining), which are associated with SARDs. We determined if machine learning could accurately differentiate between AC-2 and SARD-related AC-4/AC-30 patterns.
Methods 13,671 ANA images from SLE patients enrolled in the Systemic Lupus International Collaborating Clinics Inception Cohort (SLICC, n=2,825 images), non-SLE subjects enrolled in the Ontario Health Study (OHS, n=10,639 images), and the International Consensus on ANA Patterns (ICAP, n=207 images) were analyzed. All SLICC and OHS ANA were performed in one central laboratory using IFA on HEp-2 cells (NovaLite, Werfen, SD) and read on a digital IFA microscope (NovaView, Werfen, SD). A lab technologist (HH) with >30 years of experience identified AC-2, AC-4, and AC-30 images. Images were resized to 224x224 pixels. Three machine learning models (ANA Reader©) using a convolutional neural network (CNN) and an image feature extractor were developed to differentiate AC-2 from the other patterns. We also merged the outputs of all 3 CNNs to create a combined ANA Reader© model. 80% of the images were used for training and 20% for validation. We compared the performance of the 4 machine learning models (lab technologist as the reference standard) to determine the best prediction model.
Results The lab technologist identified 308 AC-2, 957 AC-4, and 379 AC-30 images. All 4 models performed similarly with high area-under-the-curve (AUC) scores ranging from 96.5%-97.1% (Table 1). When comparing other performance metrics, the combined ANA Reader© model performed the best with the highest accuracy (93.0%), precision (92.7%), specificity (93.2%), and F1 score (92.7%). It was tied with another CNN model (Model 2) for the second most sensitive model (92.7%).

Background/Purpose We previously evaluated hydroxychloroquine (HCQ) tapering/cessation and risk of systemic lupus erythematosus (SLE) flare in the SLICC Inception cohort. However, in our approach we did not account for potential bias due to interval-censored (IC) outcomes, where exact timing of events are unknown. Our objective was to address this, with alternative approaches to defining timing of IC events, including the Simulation Extrapolation (SIMEX) approach.
Methods We evaluated 1,543 members of the SLICC Inception cohort (January 1999 to January 2019). Adults (18+) with SLE were enrolled in this cohort within 15 months of diagnosis and followed annually with questionnaires and physician assessment. In our time-to-event analyses, time-zero was defined as cohort entry if a subject was taking HCQ at the time, or the first prescription of HCQ otherwise. HCQ tapering/cessation was defined as the first cessation or decreased dose of HCQ and modeled as a binary time-varying exposure. Multivariable proportional hazard regression assessed associations between HCQ tapering/cessation and time to SLE flare, controlling for demographics (age, sex, race/ethnicity, region, education), baseline medication (steroids, immunosuppressives, biologics), enrollment year, time between diagnosis and cohort entry, smoking status, end-stage renal disease, and body mass index. Lupus flare was defined as the earliest of: A. Increase (from prior score) of at least 4 points in the Systemic Lupus Erythematosus Disease Activity Index 2000 (SLEDAI-2K), B. Increase/initiation of SLE therapy (prednisone, immunosuppressive, or biologic), or C. SLE-related hospitalization. Since exact date of increased SLEDAI-2K was unknown, these represented IC events. We compared alternative analyses, imputing the IC event time at either the end- or the mid-point of the interval between the previous clinic visit and the visit when the outcome was reported. In sensitivity analyses we used SIMEX, a more sophisticated method based on simulations that allowed us to assess how the HR of interest changes with increasing time interval between adjacent visits. By extrapolating observed trends between the original and simulated data, we could correct the bias estimated to be within the original data, due to IC events.[1]
Results Out of the total 1,543 subjects, 398 (25.8%) decreased or stopped their HCQ at some point during their follow-up and 1,187 experienced a disease flare (76.9%). When IC event times were imputed at the end of the relevant time interval, the adjusted HR for flare related to HCQ decrease/cessation was 1.43 (95% confidence interval, CI 1.24-1.66). When IC event times were imputed at the mid-point, the point estimate for the adjusted HR was slightly higher (1.53, 95% CI1.32-1.78). SIMEX correction yielded an even higher point estimate for the adjusted HR (1.68, bootstrapped 95% CI 1.44-2.02).
Conclusions HCQ tapering/cessation was associated with greater flare risk regardless of how IC events were handled. Correcting imprecise timing of IC events tended to increase the strength of estimated associations, although confidence intervals overlapped. Limitations of these analyses include failure to account for disease status and/or other concomitant drug changes at tapering/cessation. Future analyses will address these issues (and stratify outcomes according to whether HCQ was tapered vs stopped).

Background/Purpose: Antinuclear antibody (ANA) immunofluorescence (IFA) patterns and titres are a key part of rheumatology diagnostics, however, there is considerable intra- and interlaboratory variability with manual interpretation. Replacing manual interpretation with a standardized and automated approach could help reduce variability, increasing laboratory accuracy and efficiency. We developed machine learning (ML) models (ANA Reader©) to aid laboratories in ANA pattern and titre interpretation, including a model for the nuclear dense fine-speckled (DFS) ANA pattern (AC-2), a rare pattern among systemic autoimmune rheumatic disease (SARD) patients that decreases the likelihood of these conditions.
Methods: 13,671 ANA images from SLE patients enrolled in the Systemic Lupus International Collaborating Clinics Inception Cohort (SLICC, n=2,825 images), non-SLE subjects enrolled in the Ontario Health Study (OHS, n=10,639 images), and the International Consensus on ANA Patterns (ICAP, n=207 images) were analyzed. All SLICC and OHS ANA were performed in 1 central laboratory using IFA on HEp-2 cells (NovaLite, Werfen, SD) and read on a digital IFA microscope (NovaView, Werfen, SD). As the reference standard, 1 laboratory technologist (HH) with >30 years of experience in ANA studies interpreted 13 ANA patterns and titre for each image. We developed and compared the performance of 8 ML models for ANA pattern recognition. To evaluate ANA titre, we used an ML technique for imaging processing that identified individual HEp-2 cells in the ANA images and then calculated the cell illuminance and cut-offs corresponding to each titre (1:80-1:5120). Fifty images were randomly selected to compare the titre classification based on image processing with the lab technologist as the reference standard.
Results: 6,307 images containing at least 1 ANA pattern (≥1:80) from SLICC (n=2,806 images), OHS (n=3339 images), and ICAP (n=162 images) were included. We identified 1 ML model (ANA Reader©) with the best performance for ANA pattern identification compared to the reference with a high area under the curve (AUC) score of 83.4%, modest weighted accuracy of 68.4%, precision of 67.1%, sensitivity of 70.1%, and F1 score of 67.2%. The AUC for individual ANA patterns ranged from 0.71 to 0.97 (Figure 1). There was a strong correlation between titres reported by the ANA Reader© and the technologist’s interpretation (Spearman rank 0.93, p <0.0001), where the titres reported were identical or differed Figure 1. Area under the curve (AUC) scores for the 13 antinuclear antibody (ANA) patterns using the ANA Reader© model, which had the best performance compared to 7 other machine learning techniques. by only 1 dilution in most cases (96.0%). The ANA patterns with the best performance were centromere (AUC 0.97) and pleomorphic patterns (AUC 0.97). On average, there were 5 images per patient sample for SLICC, 3 images per patient sample for OHS, and 1 image per patient from ICAP. 80% of the images were used for model training and the remaining 20% for validation. In total, there were 512 patients in the SLICC cohort, 3,559 individuals in the OHS cohort, and 207 patients from ICAP who were included in the study. Conclusions: ML has the potential to become a highly effective and efficient approach to evaluating ANA patterns and titres. The performance of our ANA Reader© is expected to improve as we continue to train our models with more ANA images. Future external validation studies and the development of other ML models to predict more complex and multiple ANA patterns and titres are also underway.

Background/Purpose We described the direct healthcare costs associated with damage accrual in patients in the Systemic Lupus International Collaborating Clinics (SLICC) Inception Cohort.[1] However, our estimates only included partial direct costs and indirect costs from lost productivity were not included. We supplemented our primary data by querying a cohort subset on all healthcare use and lost time in paid/unpaid labor and provide estimates of complete direct and indirect costs for the full cohort, stratified by damage.
Methods Between 1999 and 2011, SLE patients from 31 centers in 10 countries were enrolled into the SLICC Inception Cohort within 15 months of diagnosis and data on disease damage (SLICC/ACR Damage Index [SDI]) and limited healthcare use (ie, hospitalizations, medications, and dialysis) were collected annually through to July 2022. Starting in 2015, 18 sites collected supplemental economic data annually (ie, visits to physicians, nonphysician healthcare professionals, and the emergency room, laboratory tests, radiological/other diagnostic procedures, outpatient surgeries, help obtaining medical care, and lost time in paid/unpaid labor). Direct costs were calculated by multiplying each health resource by its corresponding 2023 Canadian unit cost. Total indirect costs included: 1) absenteeism (time lost from paid labor because of illness), 2) presenteeism (degree of patient self-reported productivity impairment in paid/unpaid labor, based on a visual analog scale), and 3) opportunity costs (additional time patients would be working in paid/unpaid labor if not ill). Opportunity costs were calculated as the difference between the time patients reported working vs that worked by an age, sex, and geographic-matched general population in paid/unpaid labor. Indirect costs from paid/unpaid labor were valued using age-and-sex-specific wages from Statistics Canada. Multiple imputation was used to predict missing cost values for the patients in the full cohort who provided only utilization data for hospitalizations, medications, and dialysis for all observations. At each assessment, patients were assigned to one of 6 damage states (ie, SDI = 0, 1, 2, 3, 4, ≥ 5) and annual costs, both unimputed and including imputations, were stratified by SDI score. Means and 95% confidence intervals were computed and compared.
Results 1694 patients (88.8% female, 48.9% White, mean age at diagnosis 34.6 years, mean disease duration at cohort enrollment 0.5 years), were followed for a mean of 10.5 (SD 5.3) years. Of these 1694 patients, 766 (89.7% female, 41.4% White, mean age at diagnosis 33.0 years, mean disease duration at cohort enrollment 0.4 years) completed the supplemental economic questionnaire. Their mean disease duration at the time of introduction of the supplemental questionnaire was 10.9 (range 3.9-19.5) years and this cohort subset provided this additional economic data for a mean of 3.5 (SD 1.9) years. Among the cohort subset completing the supplemental economic questionnaire, on average, indirect costs, primarily from unpaid labor, accounted for 81.1% of total costs (Table 1). For the full cohort, annual direct and indirect costs increased with increasing SDI (SDI=0: total costs $33,812 [95% CI $31,088, $36,537]; SDI ≥ 5: total costs $90,839 [95% CI $82,275, $99,403]) (Table 2).