Improving outcomes of pediatric lupus care delivery with provider goal setting activities and multidisciplinary care models

Abstract

Objectives:

Using an electronic health record (EHR)-enabled pediatric lupus registry, we evaluated high-quality care delivery in the context of provider goal-setting activities and a multidisciplinary care model. We then determined associations between care quality and prednisone use among youth with systemic lupus erythematosus (SLE).

Methods:

We implemented standardized EHR documentation tools to auto-populate a SLE registry. We compared pediatric lupus care index (p-LuCI) performance (range 0.0–1.0; 1.0 representing perfect metric adherence) and timely follow-up a) before vs. during provider goal-setting activities and population management, and b) in multidisciplinary lupus nephritis vs. rheumatology clinic. We estimated associations between p-LuCI and subsequent prednisone use, adjusted for time, current medication, disease activity, clinical features, and social determinants of health.

Results:

We analyzed 830 visits by 110 patients (median 7 visits/patient [IQR 4–10]) over 3.5 years. The provider-directed activity was associated with improved p-LuCI performance (adjusted β 0.05, 95%CI [0.01–0.09]; mean 0.74 vs. 0.69). Patients with nephritis in multidisciplinary clinic had higher p-LuCI (adjusted β 0.06, 95%CI [0.02–0.10]) and likelihood of timely follow-up than those in rheumatology (adjusted RR 1.27, 95%CI [1.02–1.57]). p-LuCI ≥0.50 was associated with 0.72-fold lower adjusted risk of subsequent prednisone use (95%CI [0.53–0.93]). Minoritized race, public insurance, and living in areas with greater social vulnerability were not associated with reduced care quality or follow-up, but public insurance was associated with higher risk of prednisone use.

Conclusion:

Greater attention to quality metrics associates with better outcomes in childhood SLE. Multidisciplinary care models with population management may additionally facilitate equitable care delivery.

There is a need to identify strategies to improve outcomes of children with pediatric-onset systemic lupus erythematosus (pSLE) and related conditions and ensure equitable care delivery. In adults with SLE, delivery of recommended care processes has been associated with better outcomes, including lower damage accrual (1). However, for youth with pSLE, considerable variation in care process completion exists (2,3), and literature on methods to standardize and evaluate adherence to care processes in the pediatric setting remains sparse (4). Our center has previously developed a composite index of 13 recommended pSLE care metrics, the pediatric Lupus Care Index (p-LuCI), to assess care quality in pediatric SLE across three domains: clinical assessment, comorbidity management, and population management (5). Using the p-LuCI, we identified provider-level variation in performance as well as areas in need of practice-level improvement, which informed the design of a Maintenance of Certification (MOC) activity to improve p-LuCI performance using goal setting activities and self-evaluations. At the same time, we established a multidisciplinary care model for youth with lupus nephritis with a separate population management strategy.

One of the major challenges of evaluating programmatic interventions is that data collection methods commonly employed for research and quality improvement efforts are labor intensive and often unsustainable (6). In an effort to address this challenge, we developed lupus-specific electronic health record (EHR)-enabled tools to standardize clinical documentation with embedded discrete data that could be used to auto-populate an observational pSLE research registry. The tools were designed to overcome limitations of billing-code and prescription databases by capturing clinical disease manifestations, disease activity measures, and provider-documented medication instructions within the context of routine clinical workflow.

The objectives of this study were to leverage an EHR-enabled pSLE research registry to 1) evaluate the effectiveness of two different programmatic changes – an MOC activity and a multidisciplinary lupus nephritis care model – to improve high-quality care delivery; and 2) determine whether p-LuCI performance is associated with relevant clinical outcomes. We hypothesized that provider self-evaluation and goal setting can improve p-LuCI performance, while population management strategies help ensure timely follow-up. We also hypothesized that higher p-LuCI scores are associated with reduced likelihood of any prednisone use, a frequent cause of treatment-related morbidity, among children with pSLE.

METHODS

Study Design:

This was a retrospective analysis of a prospective observational database of youth with SLE and mixed connective tissue disease (MCTD) followed at our tertiary care pediatric center. An exemption for secondary use of clinical data and waiver of informed consent was granted by the Institutional Review Board (IRB 19–016207).

Data source:

We extracted data from our EHR-based pSLE research registry from December 2018–July 2022. In December 2018, we implemented electronic health record (EHR) tools at our center to standardize documentation of patient-level pSLE manifestations and treatment history (lupus history form), as well as visit-level data for each clinical encounter (lupus visit form), including medication instructions, disease activity, disease damage and target assessments. Discrete data elements were embedded into the standardized documentation tools, which auto-populated a quality improvement dashboard and the pSLE research registry. The research registry additionally interfaced with an EHR-based steroid registry and billing data on hospital and rheumatology visit encounters in real-time (Supplemental Figure 1). The EHRbased steroid registry uses an internally validated algorithm to determine when patients first met criteria for chronic glucocorticoid prescriptions (≥15 days) and when at least 18 months had elapsed since the last active prescription (7).

Setting:

Coinciding with the development of EHR-based tools for pSLE, our center established a Lupus Program in 2018 with input from providers, patients and families, and representatives from patient advocacy groups. As of September 2018, programmatic components included a multidisciplinary lupus nephritis clinic with two rheumatologists, two nephrologists, a dedicated social worker, and a psychologist. Each month, the multidisciplinary team met to discuss population management, including outreach to patients in need of follow up visits. To ensure delivery of high-quality care for all patients with pSLE and mixed connective tissue disease (MCTD) seen in rheumatology clinic, we began an MOC activity in July 2020 to improve performance on metrics in the previously published p-LuCI (Figure 1), which is a composite measure of 13 quality indicators across three domains (standardized clinical assessment, comorbidity assessment and prevention, and population management) (5). Components of the MOC activity included a population management strategy, self-directed evaluation for individual providers, and goal-setting activities. Clinicians reviewed their p-LuCI performance, identified opportunities for improvement, and set performance goals for the next three months (Supplemental Figure 2). To obtain MOC credit, three goal-setting activities were required. Four out of 11 providers completed two activities, 4/11 completed three activities, and 3/11 completed four activities over 12 months.

Figure 1.

Timeline of interventions, beginning from the initial establishment of a multidisciplinary lupus nephritis clinic and implementation of electronic health record (EHR) documentation templates to standardize lupus clinical assessments at each rheumatology visit and patient-level summaries of disease and treatment histories. A set of quality indicators for pediatric lupus care delivery were developed in 2019 focusing on three domains: clinical assessment, comorbidity prevention and population management. A maintenance of certification (MOC) activity was subsequently initiated at our center in July 2020 to improve performance on these quality indicators.

The baseline visit for each patient was defined as the first visit occurring any time after EHR tool implementation in December 2018. Index visits for both pre-MOC (December 2018-July 2020) and MOC activity periods (July 2020-July 2022) were defined as the first visit for each patient occurring in the corresponding activity period. Patients with fewer than two outpatient rheumatology visits during the observation period were excluded from analysis (Supplementary Appendix).

Measures:

The primary outcome for high-quality care delivery was total p-LuCI score (5), modified to exclude influenza vaccination (due to incomplete source documentation of influenza vaccinations outside the health system) and documentation of adrenal insufficiency in the problem list from the numerator and denominator (due to non-rheumatology providers managing the problem list and challenges resolving timestamps). Thus, the denominator for the modified p-LuCI was 11 for patients with a diagnosis of SLE and 9 for those with MCTD, where completion of 11/11 or 9/9 metrics (100% adherence) constituted a p-LuCI of 1.0, respectively. As a secondary outcome, we defined timely follow-up care as <120 days between clinic visits to evaluate the population management components of the interventions. The primary exposure was the MOC activity (before vs. during the MOC activity period). For the subgroup of patients with lupus nephritis, we also evaluated associations between exposure to the multidisciplinary clinic model and care delivery outcomes. To evaluate longitudinal relationships between p-LuCI and improved clinical outcomes, namely a lower likelihood of any glucocorticoid requirement, we assessed prednisone use at each subsequent visit as a binary outcome (started/continued vs. not prescribed/discontinued). Prednisone use was determined by discrete, provider-entered prednisone instructions embedded in the lupus visit form when available or otherwise determined using the prescription-based EHR steroid registry (Supplementary Appendix).

Covariates:

Patient-level factors included age at the baseline visit, sex, race and ethnicity (as reported in the medical record), social vulnerability index (SVI) derived from census tract codes, and insurance type. Due to the known spatial polarization of Black neighborhoods as well as Hispanic neighborhoods (irrespective of race) in Philadelphia where our center was located (8), we analyzed mutually exclusive race and ethnicity categories as follows: Asian alone or in combination, Black or African American alone or in combination, Hispanic ethnicity with any other race, Non-Hispanic Other/unknown race, and Non-Hispanic White race. We also considered major organ manifestations, including history of nephritis, central nervous system involvement, and serositis. Visit-level time-varying factors considered included recent disease diagnosis (duration <6 months), prednisone use, use of disease-modifying antirheumatic drugs (DMARDs), and the Systemic Lupus Erythematosus Disease Activity Index 2000 (SLEDAI-2K) categorized into active disease (SLEDAI-2K >4), low/inactive disease (SLEDAI-2K ≤4), or not assessed (9). Rules applied for calculating SLEDAI scores in the setting of missing SLEDAI components are described in the Supplementary Appendix.

Statistical Analysis:

Characteristics of patients at their index visit before vs. during the MOC activity were compared using standard descriptive statistics, including Chi-square or Fisher’s exact tests for categorical variables and Student’s t-tests or Wilcoxon rank sum tests for continuous variables, as appropriate.

We used linear mixed effects models to estimate differences in continuous outcomes and modified (robust) Poisson models to estimate relative risks for binary outcomes, adjusted for time since the baseline visit, patient-level factors (age, sex, race and ethnicity, SVI, insurance, major organ manifestations), visit-level factors (disease duration <6 months from diagnosis, current prednisone use), as well as within-subject random effects. We also ran separate models restricted to SLE patients, additionally adjusted for disease activity and/or DMARD use. Separate subgroup analyses were conducted in patients with lupus nephritis to estimate differences in outcomes by clinic setting (multidisciplinary vs. rheumatology only).

We considered p-LuCI quartiles as potential cut points for predicting subsequent prednisone use, and model performance was compared using Akaike and Bayesian information criteria to select a minimum threshold at which care quality may associate with differential outcomes. We conducted a sensitivity analysis limited to visits for which provider-entered medication instructions were available compared with results including prescription-based registry data. To address potential non-random missingness, we simulated the potential range of point estimates if prednisone use or non-use was assumed for all visits missing prednisone use data. All analyses were conducted using Stata 16.0 (College Station, TX: StataCorp LLC).

RESULTS

Of 133 pSLE or mixed connective tissue disease (MCTD) cases in the EHR-based pSLE registry, 128 had both patient-level and visit-level form data available. 110 patients had at least two outpatient rheumatology visits during the observation period, 74 of which had follow-up extending through both pre-MOC and MOC periods. There was a median of 7 visits per subject [IQR 4–10], comprising a total of 830 outpatient rheumatology visits and 720 follow-up intervals. The standardized EHR documentation tool was used in 79% of visits before the MOC activity and 87% of visits during the MOC activity. Data contributing to both SLEDAI scores and medication usage was captured for 76% of pre-MOC visits and 80% of visits during MOC.

Evaluating performance in high-quality care delivery during an MOC activity

Demographic and clinical characteristics at the index visits in both pre-MOC and MOC activity periods were stable over both periods as shown in Table 1. Approximately half of patients were publicly insured, over a quarter lived in neighborhoods with the highest social vulnerability, 39% were Black, 20% were Asian, and 11–13% reported Hispanic ethnicity. A third had been diagnosed with lupus nephritis. Hydroxychloroquine use was nearly universal, and a majority (72–73%) of patients had been treated with mycophenolate.

Table 1.

Patient characteristics at index visits before and during goal-setting activity

	Pre-MOC (N=88)	MOC (N=96*)	p-value
Disease type
SLE	76 (86%)	84 (88%)	0.82
MCTD/overlap syndrome	12 (14%)	12 (12%)
Age at SLE diagnosis (y), mean (SD)	13.4 (3.5)	13.4 (3.4)	0.94
Age at index visit	16.2 (3.1)	16.6 (3.4)	0.37
Female sex, n (%)	70 (80%)	79 (82%)	0.64
Disease duration (y), median [IQR]	1.8 [0.3–4.7]	2.6 [0.3–4.8]	0.34
Race
Asian alone or in combination^	17 (20%)	19 (20%)	0.84
Black alone or in combination^	34 (39%)	37 (39%)
Other	10 (11%)	15 (16%)
White	26 (30%)	24 (25%)
Unknown	0 (0%)	1 (1%)
Hispanic ethnicity	10 (11%)	12 (13%)	0.83
Insurance
Public (Medicaid)	45 (52%)	50 (52%)	0.86
Private	40 (46%)	45 (47%)
Self-Pay/Other/Uninsured	2 ( 2%)	1 ( 1%)
Social Vulnerability Index
Lowest	22 (25%)	29 (30%)	0.87
Medium Low	19 (22%)	19 (20%)
Medium High	23 (26%)	23 (24%)
Highest	23 (26%)	25 (26%)
Unknown	1 (1%)	0 (0%)
Multidisciplinary clinic setting at index visit	6 (7%)	12 (13%)	0.19
Historical Lupus Manifestations
Arthritis	43 (49%)	48 (51%)	0.77
Serositis	7 (8%)	8 (9%)	0.93
Nephritis	29 (33%)	30 (32%)	0.84
Neuropsychiatric	5 (6%)	5 (5%)	0.93
dsDNA antibody positive	65 (74%)	70 (76%)	0.73
Hypocomplementemia	59 (68%)	67 (73%)	0.46
Lupus Treatments (Ever Use)
Hydroxychloroquine	86 (98%)	95 (99%)	0.51
Mycophenolate	63 (72%)	70 (73%)	0.84
Methotrexate	31 (35%)	32 (33%)	0.79
Rituximab	30 (34%)	29 (30%)	0.57
Cyclophosphamide	14 (16%)	13 (14%)	0.65
Azathioprine	8 (9%)	9 (9%)	0.95
Belimumab	5 (6%)	6 (6%)	0.87
Calcineurin inhibitor	4 (5%)	5 (5%)	1.00
Disease Status and Treatment at Index Visit (Current Use)
SLEDAI-2K, median [IQR]	0 [0–4]	0 [0–4]	0.82
Any DMARD useƗ	55 (93%)	51 (88%)	0.33
Prednisone use	26 (31%)	25 (29%)	0.75
Prednisone dose in users (mg/day), median [IQR]	10 [10–30]	10 [5–30]	0.59

Comparison of individual patient characteristics at each index visit, defined as the first visit occurring in each period before (December 2018–June 2020) or during the Maintenance of Certification (MOC) activity (July 2020–July 2022), using Student’s t- or rank sum tests for continuous variables and Chi-square or Fisher’s exact tests (n<5) for categorical variables.

^*Samples not mutually exclusive; N=74 individuals had follow-up extending through both periods

^{^}Includes 3 individuals reporting multiple races (Asian category: Asian and White; Black category: Black and American Indian, Black and Asian). Asian category is inclusive of “Asian” (n=20) and “Indian” (n=2) as recorded in the medical record.

^ƗDisease-modifying antirheumatic drugs (DMARD) include mycophenolate, azathioprine, methotrexate, calcineurin inhibitors, sirolimus, belimumab

Pediatric Lupus Care Index performance over time

Median unadjusted p-LuCI scores were 0.7 [IQR 0.5–0.8] during the pre-MOC period when EHR documentation tools were available for use and 0.8 [0.6 – 0.9] during the MOC activity period (Supplemental Figure 3). On average, p-LuCI scores increased over time by 0.04 per year of follow-up (95% CI [0.01–0.07]), adjusted for sociodemographic characteristics, nephritis, neurologic involvement, disease duration, and prednisone use. The MOC activity period was additionally associated with a modest but statistically significant 0.05 unit average increase in p-LuCI scores (95%CI [0.01–0.09]; marginal mean 0.74 vs. 0.69). Insurance status, race and ethnicity, and SVI were not significantly associated with p-LuCI scores on either unadjusted or adjusted analyses (Table 2). A history of nephritis and current prednisone use were associated with higher average p-LuCI scores, while visits occurring within 6 months of initial diagnosis were associated with lower scores.

Table 2.

Factors associated with p-LuCI over time

	Unadjusted*			Adjusted
	β	[95% CI]	p	β	[95% CI]	p
Month of follow-up	0.007	[0.006, 0.008]	<0.01	0.003	[0.001, 0.006]	<0.01
MOC activity period	0.04	[−0.01, 0.08]	0.12	0.05	[0.01, 0.09]	0.04
Age at baseline visit	0.002	[−0.01, 0.01]	0.72	−0.01	[−0.02, 0.01]	0.24
Male sex	0.06	[−0.03, 0.15]	0.19	0.02	[−0.06, 0.10]	0.60
Race and ethnicity
Asian alone or in combination	0.03	[−0.07, 0.14]	0.56	0.04	[−0.05, 0.13]	0.36
Black alone or in combination	0.06	[−0.03, 0.15]	0.18	0.04	[−0.05, 0.12]	0.38
Hispanic, Other/White race	0.07	[−0.05, 0.19]	0.23	0.07	[−0.04, 0.17]	0.20
Other/unknown race, Non-
Hispanic	0.02	[−0.12, 0.16]	0.77	0.02	[−0.10, 0.14]	0.80
White, Non-Hispanic	(reference)
Social Vulnerability Index
Lowest	(reference)
Medium low	0.01	[−0.09, 0.11]	0.88	0.02	[−0.06, 0.11]	0.56
Medium high	0.02	[−0.08, 0.11]	0.73	0.02	[−0.07, 0.11]	0.68
Highest	0.04	[−0.05, 0.13]	0.36	0.06	[−0.04, 0.15]	0.23
Insurance
Private	(reference)
Medicaid	0.03	[−0.04, 0.1]	0.36	−0.02	[−0.09, 0.04]	0.48
Self-Pay/Uninsured	0.10	[−0.12, 0.32]	0.38	0.02	[−0.17, 0.20]	0.87
History of nephritis	0.16	[0.09, 0.22]	<0.01	0.08	[0.02, 0.14]	0.01
History of neurologic involvement	0.12	[−0.03, 0.27]	0.11	0.07	[−0.05, 0.19]	0.24
Recent diagnosis (<6 months)^	−0.09	[−0.14, −0.05]	<0.01	−0.12	[−0.17, −0.08]	<0.01
Current prednisone use^	0.14	[0.1, 0.18]	<0.01	0.13	[0.09, 0.18]	<0.01

Mixed effects linear regression with patient-level random intercept (N=683 visits for 107 unique patients). Three patients and 37 visits were dropped due to missing data for one or more demographic characteristics (1 patient) or prednisone use (34 visits).

^*All univariable analyses are additionally adjusted for time (month of follow-up)

^{^}Time-varying covariates

During the MOC activity period, improvements in p-LuCI were driven by completion of clinical assessments (SLICC damage index, physician global, target attestation, lupus characteristics review, as well as comorbidity assessment [blood pressure, lipid and vitamin D]) (Figure 2).

Figure 2.

Bar graph representing the total proportion of visits at which each quality metric was met, both before (hatched blue) and during (solid orange) the maintenance of certification (MOC) activity. * p<0.05, unadjusted; † Denominator excludes visits for patients with MCTD; ‡ Denominator limited to visits meeting criteria for chronic steroid use

Timely outpatient rheumatology follow-up

There was no significant increase in timely follow-up during the MOC activity vs. pre-MOC period in adjusted models (67% vs. 61%, adjusted RR 1.10, 95% CI [0.94–1.29]) (Table 3). Upon restricting the analysis to patients with SLE, additional adjustment for SLEDAI scores and DMARD use yielded similar results (65% vs. 60%, RR 1.07, 95% CI [0.90–1.28]) (Supplemental Table 1). Hispanic ethnicity, younger age, disease duration <6 months, and current prednisone use were associated with a higher likelihood of timely follow-up (Table 3). Of note, higher SVI, Black race, and insurance status were not associated with a lower likelihood of timely follow-up during the observation period.

Table 3.

Factors associated with timely outpatient rheumatology follow-up

	Adjusted RR	[95% CI]	p-value
Month of follow-up	0.99	[0.98, 1.00]	0.02
MOC activity period	1.10	[0.94, 1.29]	0.24
Age at baseline visit (years)	0.96	[0.93, 0.99]	0.01
Male sex	0.93	[0.74, 1.17]	0.55
Race and ethnicity
Asian alone or in combination	1.07	[0.81, 1.41]	0.64
Black alone or in combination	1.21	[0.91, 1.61]	0.20
Hispanic White/Other	1.34	[1.04, 1.72]	0.02
Other/Unknown race, Non-Hispanic	0.63	[0.36, 1.12]	0.12
White, Non-Hispanic (reference)	-
Social Vulnerability Index
Lowest (reference)	-
Medium low	1.07	[0.81, 1.42]	0.64
Medium high	0.95	[0.69, 1.29]	0.73
Highest	1.06	[0.81, 1.38]	0.67
Insurance
Medicaid	1.06	[0.87, 1.28]	0.58
Self-Pay/Uninsured	1.10	[0.88, 1.38]	0.41
History of nephritis	1.14	[0.97, 1.33]	0.11
History of neurologic manifestations	1.03	[0.76, 1.38]	0.87
Recent diagnosis (<6 months)*	1.29	[1.09, 1.54]	0.00
Current prednisone use*	1.18	[1.01, 1.39]	0.04

Estimates from modified robust Poisson models with subject-level random effects (N=107 SLE/MCTD patients; 683/730 visits with complete data).

^*Time-varying covariate

Lupus nephritis care quality in the context of a multidisciplinary care model

In the subgroup of 35 patients with lupus nephritis (comprising a total of 252 visits), we similarly observed a 0.07 unit adjusted increase in p-LuCI associated with the MOC activity period (95% CI [0.02–0.12], p=0.01).

A total of 19 patients with lupus nephritis (of which 47% had proliferative disease, 37% pure membranous) were evaluated in the multidisciplinary clinic at least once during the observation period; the remaining 16 patients (69% proliferative, 19% membranous) were followed exclusively in general rheumatology clinic. Those evaluated in multidisciplinary clinic were on average younger (mean age 14.7 [SD 3.5] vs. 17.1 [2.7], p=0.03), with higher disease activity at the baseline visit (median SLEDAI-2K score 6 [2–11] vs. 0 [0–2], p<0.01), and the majority were publicly insured (63% vs. 25%, p=0.06). The median p-LuCI at each multidisciplinary clinic visit was 0.91 [0.82–1.00] (n=118) compared to 0.81 [0.60–0.90] (n=134) for rheumatology clinic visits (p<0.01, unadjusted). On average, multidisciplinary visits were associated with a 0.06 higher p-LuCI compared to rheumatology clinic visits (95% CI [0.02–0.10], p=0.01; marginal mean 0.87 vs. 0.82), adjusted for time, sociodemographic factors, disease duration, disease activity, and prednisone use. Furthermore, patients who received care in the multidisciplinary clinic at any time during follow-up had a 0.10 higher average p-LuCI compared to those receiving care exclusively in the rheumatology clinic (95% CI [0.03–0.16], p<0.01; marginal mean 0.88 vs. 0.78). This was driven by greater completion of disease characteristics review, pneumococcal vaccination, assessment of disease activity and disease damage, as well as follow-up within 180 days (data not shown).

As with the full cohort, there was no statistically significant increase in timely outpatient follow-up within 120 days for the lupus nephritis subgroup during the MOC activity vs. pre-MOC periods (77% vs. 65%, adjusted RR 1.17, [0.86–1.61]). However, being seen in the multidisciplinary clinic was associated with significantly higher likelihood of timely follow-up compared to being seen in rheumatology clinic (adjusted RR 1.27, 95% CI [1.02–1.57], p=0.03). Black race and Hispanic ethnicity were also independently associated with greater likelihood of timely follow-up compared to non-Hispanic White race (adjusted RR 1.50, 95% CI [1.11, 2.02], p=0.01 and 1.37 [1.02, 1.84], p=0.03, respectively). Moreover, patients with lupus nephritis living in census tracts with medium-high social vulnerability were also 1.45 times more likely to have timely follow-up compared to those from census tracts with the lowest SVI (95% CI [1.01, 2.08], p=0.05) (Supplemental Table 2).

Association between high-quality care delivery and clinical outcomes

Medication instructions were entered by providers in the lupus visit form for 106/110 patients for a median of 90% of visits per patient during the observation period [IQR 73–100%], corresponding to a visit-level completion rate of 669/830 (81%). Current use of prednisone was documented in 279/669 (41.7%) visits with any medication template entry, and a discrete prednisone dose was also entered for 254/279 (91%) instances of documented prednisone use. For the remaining 161 visits without provider-entered medication instructions, active (n=23) vs. inactive (n=99) prednisone prescriptions were obtained from the EHR-based steroid registry for 122/161 (76%) visits, yielding 39/830 (5%) visits for which prednisone use data was missing. No patients with nephritis were missing prednisone use data. Shorter follow-up time and lower p-LuCI performance were associated with missing prednisone data (Supplemental Table 3).

In the total SLE/MCTD cohort, a p-LuCI ≥0.5 vs. <0.5 (lower quartile) was associated with an adjusted RR of 0.62 (95% CI [0.48–0.82], p<0.01) for subsequent prednisone use at the next visit, adjusted for current prednisone use, disease duration, baseline major organ involvement, and sociodemographic factors. Upon restricting the analysis to SLE patients only and further adjusting for current disease activity, p-LuCI ≥0.5 was associated with a 0.72-fold lower risk of subsequent prednisone use (95% CI [0.53–0.93], p=0.01). Compared to privately insured patients, those with public insurance were 1.4 times more likely to require prednisone at the subsequent visit (Table 4). In contrast, medium-high SVI was significantly associated with a 0.66-fold reduction in risk of prednisone use. In addition, the lack of SLEDAI score assessment at each visit was also independently associated with a 1.25-fold [1.04–1.49] and 1.42-fold [1.16–1.73] higher likelihood of subsequent prednisone use compared to an assessment of active disease (SLEDAI >4) or low disease activity (SLEDAI ≤4), respectively. Results were similar in a sensitivity analysis limited to visits in which provider-entered prednisone use instructions were available (adjusted RR 0.70 for pLUCI≥0.5 vs. <0.5, 95% CI [0.50–0.98], p=0.04). Assuming prednisone use at all visits missing prednisone data vs. non-use yielded a range of possible RRs from 0.82 [0.63–1.07] to 0.68 [0.51–0.92], respectively.

Table 4.

Association between p-LuCI performance and subsequent prednisone use

	RR	[95% CI]	p-value
Month of follow-up	1.00	[1.00 – 1.01]	0.12
p-LuCI ≥0.5*	0.72	[0.54 – 0.96]	0.03
Age at baseline visit	1.01	[0.98 – 1.04]	0.49
Male sex	1.25	[0.97 – 1.60]	0.08
Social Vulnerability Index
Lowest	(reference)
Medium low	1.04	[0.81 – 1.33]	0.75
Medium high	0.66	[0.48 – 0.90]	0.01
Highest	0.85	[0.69 – 1.05]	0.14
Race and ethnicity
Asian alone or in combination	0.98	[0.73 – 1.33]	0.92
Black alone or in combination	1.20	[0.91 – 1.60]	0.20
Hispanic, Other/White race	1.23	[0.93 – 1.63]	0.15
Other/Unknown race, Non-Hispanic	1.38	[0.97 – 1.98]	0.08
White, Non-Hispanic	(reference)
Insurance
Medicaid	1.42	[1.17 – 1.72]	<0.01
Self-Pay/Uninsured	0.94	[0.61 – 1.44]	0.77
History of nephritis	1.13	[0.91 – 1.40]	0.26
History of neurologic manifestations	1.45	[1.02 – 2.06]	0.04
History of serositis	1.21	[1.00 – 1.47]	0.05
Recent diagnosis (<6 months)*	1.40	[1.19 – 1.64]	0.00
Current prednisone use*	9.36	[6.12 – 14.31]	0.00
Current disease activity*
SLEDAI-2K >4	(reference)
SLEDAI-2K ≤4	0.88	[0.75 – 1.04]	0.14
SLEDAI-2K not assessed	1.25	[1.04 – 1.49]	0.02

Factors associated with any prednisone use at each subsequent visit in 92 patients with SLE only (N=591 visits).

p-LuCI = pediatric Lupus Care Index (range 0.0–1.0); SLEDAI = Systemic Lupus Erythematosus Disease Activity Index 2000

^*Time-varying covariate

DISCUSSION

At our center, uptake of pSLE-specific EHR documentation tools was high in the context of interventions to improve high quality care. Without any manual chart abstraction, we were able to evaluate changes over time in care quality performance and clinical outcomes in the setting of two programmatic changes. We observed small but statistically significant improvements in care quality metric performance over time with a relatively low-intensity intervention involving provider self-directed goal-setting activities. At the same time, a complex intervention involving a multidisciplinary care model for patients with lupus nephritis was associated with higher care quality performance and more timely follow-up care. Importantly, in the context of these interventions, there were no significant disparities in care quality metrics by race and ethnicity, insurance status or neighborhood-level social vulnerability indices. Finally, we also demonstrate a longitudinal relationship between high-quality care delivery and reduced likelihood of any prednisone use, suggesting that improving adherence to care quality metrics may also improve clinical outcomes.

While there is a significant body of literature describing care process measures for SLE, very few studies have been able to address how improving recommended care processes translates to better clinical outcomes (1). A previous study in children with SLE did not find that adherence to individual quality metrics associates with decreased damage (3). However, by using a composite index of care quality, we demonstrate that adherence to at least 50% of care quality indicators is associated with significant reductions in the risk of subsequent prednisone use. Interestingly, the association between lack of SLEDAI assessment at a given visit and a higher likelihood of subsequent prednisone use raises the question of whether even more frequent disease activity assessment, such as in a treat-to-target framework (10), could further reduce prednisone exposure and related toxicity. As a composite measure, p-LuCI performance represents overall attention to best practices; therefore, the relationship between p-LuCI scores and prednisone use may be driven by factors other than the individual care processes. Of note, public insurance remained associated with a higher risk of subsequent prednisone use, independent of care quality. This suggests that either the p-LuCI does not fully capture processes driving prednisone use, or that systems and individual-level factors mediate the relationship between insurance status and prednisone use. Studies have demonstrated that Black, Asian and Hispanic adults with SLE receive higher maximum prednisone doses and over longer periods of time compared to their White counterparts (11,12). Frequent outpatient care was protective, with Black and Hispanic individuals having fewer visits (12). In our cohort, there were no significant disparities in timely rheumatology follow-up, which may have mitigated differences in prednisone use by race and ethnicity but does not explain differences by insurance status. It is possible that delays in initial access to subspecialty care or use of steroid-sparing agents drive this difference. Qualitative research may elucidate reasons for higher prednisone exposure among publicly insured children with SLE (13).

Measurement of health care quality has been proposed as an important mechanism through which disparities in care can be identified and ameliorated (14). We did not observe racial or socioeconomic disparities in quality metric performance during the study period, which spanned introduction of a dedicated lupus social worker, the multidisciplinary lupus nephritis clinic, the MOC activities, as well as the coronavirus disease 2019 (COVID-19) pandemic. This is notable, particularly as the pandemic both exposed and exacerbated existing disparities in many other contexts (15,16), and was ongoing throughout the MOC activity, potentially even attenuating improvements in care delivery outcomes. Furthermore, living in areas with medium-high social vulnerability was unexpectedly associated with a lower risk of subsequent prednisone use compared to living in areas with the lowest social vulnerability, which may reflect efforts to ensure timely follow-up care, particularly among patients with lupus nephritis with greater prednisone exposure. While we cannot determine which of the complex interventions contributed to maintaining equitable care delivery, our data suggest that interventions combining social work support with standardized clinical assessment and/or multidisciplinary care models to eliminate health disparities should be systematically studied. Hybrid effectiveness-implementation designs may be particularly well suited to the evaluation of these types of complex interventions (17).

While multidisciplinary care models are considered standard of care in oncology and have attained recognition at policy levels (18,19), published literature on the implementation of multidisciplinary care models for SLE remains sparse. One study demonstrated that a multidisciplinary care model for adults with lupus nephritis was associated with decreased time to kidney biopsy and improved performance on select quality metrics (20). In our study, we similarly observed better p-LuCI performance associated with a multidisciplinary care model for pediatric lupus nephritis, particularly with respect to routine assessment of disease activity and damage, which is a key component of treat-to-target approaches (10). As being seen in dedicated lupus clinics or by providers with higher volumes of lupus patients is also associated with higher care quality performance (21), it is not possible to directly attribute higher p-LuCI performance to the multidisciplinary components, such as collaboration, coordination, co-localization, interdisciplinary integration of knowledge, or patient-centered care (22). Nevertheless, our findings support the growing body of literature that suggests patients with lupus would benefit from multidisciplinary chronic care models that are routinely implemented for other complex chronic conditions.

In addition to higher care quality, we also observed a greater likelihood of timely follow-up for patients with lupus nephritis associated with care in the multidisciplinary clinic compared to general rheumatology clinic. Although population management was also a component of the MOC activity for all rheumatology providers, each provider could choose whether or not to set population management as a goal. In contrast, the multidisciplinary care team had a distinct population management strategy, in which the team confirmed the intended follow-up timeframe after each visit and also tracked and reviewed visit intervals on a monthly basis for all patients ever evaluated in the multidisciplinary clinic. The social worker played a key role in re-contacting families prior to each visit and evaluating transportation needs or other barriers. Albeit challenging to quantify, the psychologist likely also facilitated a bi-directional relationship between better care delivery and patient engagement by addressing mental health needs in real-time. In this context, patients belonging to historically marginalized racial groups, living in areas with greater social vulnerability, or with public insurance were equally, if not more likely, to receive timely follow-up care. This has important implications for health equity, as access to appropriate outpatient care has been associated with receipt of recommended care (23), and has also been hypothesized to mediate disparities in renal outcomes by insurance status (24,25) and by race (11). As such, population management strategies that integrate dedicated social work support may help mitigate differential access to timely rheumatology follow-up and reduce health disparities.

Strengths of our study include availability of comprehensive, longitudinal data in a pediatric lupus cohort with racial, ethnic, and socioeconomic diversity. Use of lupus-specific EHR documentation tools enabled collection of discrete data at the point of care, eliminating manual review of unstructured data. These tools can potentially be implemented across institutions using a common data model and improve upon current examples of EHR-based learning health networks by providing access to disease activity scores and clinical phenotypes (26–28). There are also disadvantages of EHR-enabled registries that present limitations to the current study. Data completeness is dependent on uptake of EHR forms, which may not be easily replicated in other settings. In addition, data programming requirements limit spread to practices with limited information technology resources. However, as these efforts align with policies regarding “meaningful use” of EHRs, investment by institutions is warranted (29). Additional limitations of our study include incomplete generalizability, as academic centers tend to have higher performance on recommended SLE care (30), also evidenced by nearly 100% adherence to hydroxychloroquine prescribing in our study. Missing data was present and could have biased our point estimates to a certain degree, albeit unlikely to change the direction of the effect. Analysis of the impact of interventions on other relevant glucocorticoid outcomes, such as cumulative prednisone dose or tapering, would have required additional chart review; albeit this data infrastructure would still substantially reduce manual review and be more accurate than electronic prescriptions (31). Due to low damage scores in our cohort and inclusion of damage assessment in the p-LuCI, we could not evaluate relationships between p-LuCI and damage accrual. However, future studies combining EHR-enabled registries with prospective data could evaluate relationships between care quality and longer term outcomes. Lastly, changes over time cannot be directly attributed to any particular intervention in observational studies. We could not account for mitigation efforts and shielding behaviors during the COVID-19 pandemic, which could have decreased rates of follow-up and metric completion both in the months prior to and during the MOC activity. Reassuringly, the trends in rates we observed remained relatively stable over the study period.

In summary, an EHR-enabled pediatric lupus registry with standardized clinical documentation enables evaluation of longitudinal disease and care delivery outcomes without manual review of medical records. In a real-world population of youth with SLE, we demonstrated improvements in delivery of high quality care associated with provider-directed goal setting activities as well as a multidisciplinary care model for patients with lupus nephritis and implementation of population management strategies. In the context of these interventions, we did not observe any racial or socioeconomic disparities in timely outpatient rheumatology care or receipt of recommended care processes. Routine, automated assessment of care processes and disease status can serve as an important means to ensure equitable care delivery and evaluate interventions designed to deliver comprehensive, patient-centered care.

Significance and Innovations.

• Provider self-directed goal setting activities can improve performance on care quality metrics for children with lupus

• Multidisciplinary care models with social work support and population management strategies are associated with better care quality and timely follow-up care

• Greater adherence to a composite index of care quality metrics for pediatric lupus is associated with reduced likelihood of any prednisone use at each subsequent visit

• Standardized documentation of lupus characteristics and clinical assessments can facilitate both measurement of high-quality care delivery as well as research