A longitudinal analysis of costs associated with change in disease activity in systemic lupus erythematosus

Abstract

Objectives:

To estimate the economic consequences of changes in disease activity on healthcare resource utilization (HRU) and costs.

Methods:

A retrospective longitudinal study of systemic lupus erythematosus (SLE) patients receiving care in a regional integrated health delivery system in the US from 01/2004 through 03/2011 was conducted using electronic health records, medical chart reviews, and claims. Eligible patients were ≥18 years old, with ≥1 rheumatologist-confirmed SLE diagnosis and ≥1 eligible rheumatology encounter. Patients were continuously enrolled ≥90 days before and ≥30 days after the encounters. Charts were manually reviewed to estimate SLEDAI scores. Average unit costs of each medical procedure, facility use, and prescription were estimated from a payer perspective (2011 USD) using a managed care claims database. HRU and costs were calculated for the 30-day period surrounding every SLEDAI score date (10 days before and 19 after). Relationships between HRU/costs and SLEDAI scores were estimated using mixed-effect models.

Results:

Overall, 178 SLE patients were included; mean age was 50.6 years, 91% were female, and 95.5% Caucasian. Patients had a total of 1343 encounters with SLEDAI scores over an average period of 1035 days. Reductions of SLEDAI scores were associated with reductions in HRU and costs. SLEDAI score reductions of 4-points were associated with reductions of 10% HRU and 14% costs over a 30-day period; reductions of 8-points had associated reductions of 19% HRU and 26% costs; and reductions of 10-points had associated reductions of 23% HRU and 31% costs. Annualized, changes in SLEDAI scores are associated with changes of $2485 (SLEDAI score change: 10–6), $4624 (10–2), and $5579 (10–0), respectively.

Conclusion:

Reductions in disease activity were associated with substantial reductions of HRU and costs.

Limitations:

Only short-term effects of disease activity change were investigated, disregarding other potential benefits of low disease activity on long-term organ damage prevention or comorbidities.

Keywords::

• Systemic Lupus Erythematosus
• SLEDAI
• Costs
• Health resource utilization

Introduction

Systemic lupus erythematosus (SLE) is a chronic autoimmune disorder that may affect multiple organ systems. SLE is a relatively uncommon disease, with incidence rates ranging from 1.8–7.6 cases per 100,000 persons, varying by age and gender1. SLE occurs in women 9-times more frequently than men, with peak onset at child-bearing age2.

SLE may cause permanent damage in multiple systems including the dermatologic, renal, cardiac, pulmonary, gastrointestinal, musculoskeletal, and central nervous systems. It impairs patients’ quality-of-life and poses substantial economic burdens in terms of direct healthcare expenditures and indirect costs associated with disability, absenteeism, lower productivity, and unemployment. A review of cost-of-illness studies on SLE estimated that the average direct healthcare costs per patient-year ranged from $3735–$14,410, with two estimates based on claims data of ∼$13,000 per year in the US3. SLE patients with organ damage, such as nephritis, incurred substantially higher costs4. In most studies, the indirect costs (e.g., sick leave, short- and long-term disability) per patient-year exceeded the direct costs and were estimated to range from $1093–$14,6145.

Despite a number of studies focusing on the overall economic burden of SLE, there is a lack of research on the costs associated with different disease severities and manifestations3. The few studies that have attempted to estimate this relationship applied cross-sectional designs. Zhu et al.4 conducted an economic evaluation of the relationship between flares in SLE patients and healthcare costs in Hong Kong from a social perspective. They estimated that the total average healthcare costs per patient-year were more than double in patients who experienced disease flares compared to patients without flares ($22,580 vs $10,870, in 2006 USD) and concluded that patients with flares incurred both higher direct and indirect costs. Flares involving renal and neuropsychiatric manifestations were found to be associated with particularly high costs4. Similarly, Sutcliffe et al.6 determined that higher disease activity was associated with higher direct and indirect costs in the UK, where disease activity was measured by a revised version of the Systemic Lupus Activity Measure (SLAM-R). A 1-point increase in the SLAM-R was estimated to be associated with an increase of 239.4 GBP (pounds sterling) in direct costs and 317.4 GBP in indirect costs (in 1996–1997 GBP). These studies, based on cross-sectional designs, showed that more severely-affected SLE patients, where severity was estimated by the SLAM-R or the presence of flares, had higher direct and indirect costs compared with patients with less severe SLE. However, little is known about the association between changes in disease activity and corresponding changes in healthcare resource utilization (HRU) and costs.

A characteristic of SLE is the extreme variability of its manifestation, both across patients and within patients over time. Clinically, higher disease activity or flares have been identified as important predictors of increased mortality risk and organ damage7. Therefore, when assessing the costs of SLE and its treatment, in addition to cross-sectional variations of SLE costs among patients, it is important to understand the temporal relationship between disease activity and costs over time. The association between healthcare costs and disease activity changes is also essential to evaluating the economic consequence of any medical or therapeutic intervention that could alter the disease activity of SLE patients over time. To address this question, this study aimed to estimate the economic consequence of the change in SLE disease activity, measured by the SLE disease activity index (SLEDAI) score, on HRU and direct costs.

Methods

Study design

A retrospective longitudinal study was conducted on SLE patients who received care between January 2004 and March 2011 in the Geisinger Health System (GHS). Analyses were conducted at the patient-encounter level, where each patient could have had multiple encounters over time.

Data source and SLEDAI score calculation

The study used data from the MedMining database of the GHS, which includes both electronic medical records (EMRs) and claims data. The EMRs include de-identified patient charts, laboratory orders and results, vital signs, medication orders and administrations, procedure orders, diagnoses, and general demographic information (e.g., age, race, gender). The EMRs document events in both inpatient and outpatient settings for primary and specialty care. All data elements are routinely captured by Geisinger at the point of care. The MedMining database also includes claims data from the Geisinger Health Plan (GHP), a managed care company with over 250,000 members and 25,000-plus contracted physicians and facilities. The GHS is a vertically integrated system that provides care to patients across numerous clinical sites and hospitals located in a 40-county region throughout the state of Pennsylvania. Geisinger’s patient population is mostly Caucasian, with a median of 15 years of continuous care within the GHS and a very low migration rate. No patient-identifiable information was collected from the chart data or insurance claims data.

Administrative claims data from GHP include comprehensive HRU records that also list medical services incurred outside of the GHS; however, financial information (e.g., total amount reimbursed) is not available. To estimate the costs from a third-party payer perspective, all claims from SLE patients in a large, aggregated US claims database (the costing sample) were used to impute costs for both the claims data and EMRs of the GHS patients. For each medical service and medical facility record in the EMR or claims data, all claim records of the same type in the costing sample were identified based on CPT code or inpatient facility code. The average costs of these claims were calculated and used as the cost for the corresponding service received by the GHS patient. Similarly, drug costs were imputed by averaging the paid amount of pharmacy claims with the same national drug code (NDC) in the costing sample. All costs were adjusted for inflation using the medical component of the CPI and are reported in 2011 USD.

Since SLEDAI scores are not routinely used and collected in clinical practice, manual chart reviews were performed by three rheumatology nurse specialists at Geisinger to calculate a SLEDAI score for each eligible encounter. These nurses have experience working in a GHS rheumatology department and were supervised by a staff rheumatologist. The SLEDAI score is based on the presence of 24 descriptors in nine organ systems, weighted according to severity. In accordance with the original SLEDAI score definition, SLEDAI scores were calculated based on the presence of descriptors observed during the 10 days prior to the eligible medical encounters. For descriptors that are based on laboratory test results, an additional 14-day period following the encounter date was reviewed, as there may be a lag between the time the test is ordered and when results become available. The nurses reviewed each patient’s EMRs during the 25-day period (i.e., 10 days before and 14 days after the medical encounter) surrounding eligible encounters, and a SLEDAI score was calculated for every corresponding period. The SLEDAI scores were calculated using the original measure as validated in the initial studies conducted in the early 1990s8.

Patient selection and medical encounters

Patients were selected from a cohort of adult SLE patients who received care between January 2004 and March 2011 in GHS and carried GHP insurance. More specifically, to be eligible, patients were at least 18 years old, had at least one rheumatologist-confirmed SLE diagnosis (ICD-9 710.0x), and had at least one eligible rheumatology encounter—defined as a visit to a rheumatology, nephrology, or emergency department associated with a diagnosis for SLE. Patients were required to have at least one eligible encounter, that is, an encounter for which the patient was continuously enrolled in the GHP for at least 90 days before and 30 days after the eligible encounter. A SLEDAI score was calculated for every eligible encounter. The date of the corresponding eligible rheumatology encounter was termed the SLEDAI score encounter date. Patients may have multiple eligible encounters and corresponding SLEDAI scores.

Outcomes

The primary outcomes were mean HRU and healthcare costs measured over a 1-month period surrounding the SLEDAI encounters. For the analysis of HRU, the number of inpatient visits and total number of inpatient days, the number of emergency room (ER) admissions, the number of other healthcare visits, and the total number of medical encounters (i.e., the sum of inpatient days, outpatient visits, ER admissions, and other visits) were analysed. Similarly, healthcare cost outcomes included medical costs (i.e., the sum of medical service costs and inpatient costs), pharmacy costs, and total healthcare costs (i.e., the sum of medical and pharmacy costs). Monthly HRU, calculated using both the EMRs and claims data, and imputed costs (measured from a payer’s perspective and reported in 2011 USD) incurred during the 30-day period surrounding each SLEDAI score encounter date—from 10 days prior to the SLEDAI score date to 19 days following the SLEDAI score date—were analysed. In addition, in order to descriptively examine the trends in HRU and costs associated with average SLEDAI scores, the total number of medical encounters and total healthcare costs were estimated over the entire continuous eligibility period for each patient. The number of medical encounters and costs were then annualized.

Statistical analyses

Patient characteristics, including age, gender, race, body mass index (BMI) for a sub-set of patients for whom the information was available, and smoking status were assessed on the date of the patients’ first SLEDAI score encounter date. Patients’ comorbidity profile was determined based on the Charlson Comorbidity Index (CCI), a composite score of overall disease burden, that was calculated based on comorbidities observed during the entire eligibility period prior to the first SLEDAI score encounter date. Patients’ SLE treatment history was assessed during the 90-day period prior to the first eligible encounter. This information was extracted from both claims data and EMRs. Means and standard deviations (SD) were reported for continuous variables, and the number of patients and percentages of the total sample were reported for categorical variables.

Descriptive statistics of the patients’ level of disease activity were also reported, including the number of days between patients’ first and last SLEDAI score encounter dates, the average number of eligible encounters per patient, the average SLEDAI score on the first scoring encounter date, and the difference between patients’ lowest and highest SLEDAI score among all their eligible encounters. In addition, the percentage of patients with their average SLEDAI score equal to 0, greater than 0, and greater than or equal to 6 were reported. Additional descriptive statistics were also reported, including the average SLEDAI score among all SLEDAI scores, and among patients with at least two eligible encounters, the average change in SLEDAI score compared to the previous SLEDAI score, and the average number of days between two eligible encounters.

Annualized HRU, measured over the entire continuous eligibility period, was reported among all patients by type of HRU, including the average and median number of medical encounters, inpatient days, ER admissions, and inpatient visits per patient. Annualized healthcare costs, also measured over the entire continuous eligibility period, including average and median total healthcare costs, medical costs, and drug costs, were reported, stratified by patients’ average SLEDAI score over the same period.

The association between SLEDAI scores and HRU and costs over the 30-day period surrounding SLEDAI score encounters was analysed using mixed effect models. These models accounted for longitudinal data of multiple observations per patient, assuming patient-level random effects. The random intercepts at the patient level allow inherent disease severity to vary across patients and also enable the estimation of the effect of disease severity to change within patients. The relationship between HRU and the SLEDAI score was estimated using a mixed effect model with a logarithmic link function and negative binomial distribution, while the relationship between healthcare costs and the SLEDAI score was estimated using a mixed effect model with a logarithmic link function and gamma distribution. For all regression models of HRU and healthcare costs, control variables included age, gender, race, CCI, and smoking status. These models estimated the percentage of change in a dependent variable (HRU or costs) associated with each unit change in the SLEDAI score. To estimate the effect of SLEDAI score changes on the original scale of the dependent variables, expected HRU and costs at each possible SLEDAI score level was calculated for each patient and averaged across the sample. The impact of the changes in SLEDAI score on HRU and costs could, therefore, be directly estimated by comparing the expected number of medical encounters or average costs between different levels of SLEDAI scores. All analyses were performed using SAS 9.2 (SAS Institute Inc., Cary, NC).

Results

Patient characteristics

A total of 178 patients met sample selection criteria and were included in this study. The majority (91.0%) were female and Caucasian (95.5%). The average age was 50.6 (SD = 15.0) and the average BMI was 28.6 (SD = 7.1). Only 15.2% of the patients were smokers. The average CCI was 1.9 (SD = 1.5). During the 90 days prior to the first eligible encounter date, 33.1% of patients used corticosteroids, 33.1% used hydroxychloroquine, 16.3% used NSAIDs, and 7.3% used immunosuppressive agents (Table 1).

Table 1. Patient characteristics.

	n = 178
Age (years; mean ± SD)	50.6 ± 15.0
Male, n (%)	16 (9.0%)
Race, n (%)
White	170 (95.5%)
Black	3 (1.7%)
Asian	2 (1.1%)
Hispanic	1 (0.6%)
Other	2 (1.1%)
Body mass index^a (mean ± SD)	28.6 ± 7.1
Smoker, n (%)	27 (15.2%)
CCI^a (mean ± SD)	1.9 ± 1.5
Use of SLE therapies^b,c
Corticosteroids	59 (33.1%)
Hydroxychloroquine	59 (33.1%)
NSAIDs	29 (16.3%)
Immunosuppressives	13 (7.3%)
Mycophenolate Mofetil	3 (1.7%)

^aMeasured during the entire eligibility period prior to the index date.

^bMeasured during 90 days prior to the first SLEDAI score.

^cCalculated using drug claims data only.

SLEDAI scores

The selected patients had a total of 1343 eligible encounters. At the patient level, the average time between the first and last eligible encounter was 1035 days (2.8 years), with an average of 7.5 SLEDAI scores reported per patient. The mean SLEDAI score on the first encounter was 2.3. Forty-five patients (25.3%) had all of their SLEDAI scores equal to 0, and 38.8% of patients had at least one SLEDAI score greater than or equal to 6. Across all eligible encounters, the average SLEDAI score was 2.15 (SD = 3.04), and the median SLEDAI score was 0; 52.9% (710) of the SLEDAI scores among all eligible encounters were equal to 0. Among the patients with at least two encounters, the average difference between the highest and the lowest SLEDAI score was 4.35 (SD = 4.09), the average change from the previous SLEDAI score was −0.06 (SD = 3.00), and the average number of days between two consecutive encounters was 128 days (SD = 149) (Table 2).

Table 2. Description of SLEDAI scores.

	Mean ± SD	n	%
Number of encounters per patient	7.54 ± 8.41
Number of days between first and last SLEDAI score^a	1035 ± 779
First SLEDAI score	2.29 ± 3.10
Difference between a patient’s highest and lowest SLEDAI score^a	4.35 ± 4.09
Number of patients with average SLEDAI score
Avg SLEDAI score = 0		45	25.3%
Avg SLEDAI score > 0		133	74.7%
Avg SLEDAI score ≥ 6		17	9.6%
Number of patients with at least one SLEDAI score
>0		133	74.7%
≥6		69	38.8%
SLEDAI scores (n = 1343)
All SLEDAI scores	2.15 ± 3.04
Change since previous SLEDAI score^a	−0.06 ± 3.00
Number of days since previous SLEDAI score^a	128 ± 149

^aAmong the 144 patients with ≥2 SLEDAI score calculation.

Annualized HRU and costs

Over their entire continuous eligibility period, the annualized average number of medical encounters per patient was 34.2 (SD = 37.0; median = 24.0), and the average annualized total healthcare cost was $18,839 (SD = $29,776; median = $9655) (Tables 3 and 4). When healthcare costs were reported per patient’s average SLEDAI score, results suggested a trend of the median annualized total healthcare costs increasing with the average SLEDAI score; however, this trend was not statistically significant for medical costs. Patients with an average SLEDAI score equal to 0 had median annualized healthcare costs of $7512; patients with an average SLEDAI score between 0–2 had median annualized healthcare costs of $9529; patients with an average SLEDAI score between 2–4 had median annualized healthcare costs of $10,655; and patients with an average SLEDAI score greater than 4 had median annualized healthcare costs of $13,756. However, the annualized average total healthcare costs did not show a clear increasing trend following increased disease activity (Table 4).

Table 3. Annualized HRU.

	Median	Mean	SD
Total HRU	24.0	34.2	37.0
Inpatient days	0.4	1.9	4.1
Outpatient visits	19.2	24.8	24.1
ER admissions	0.0	0.3	0.9
Other visits	1.5	7.1	22.3

Table 4. Annualized healthcare costs.

		Total costs			Medical costs			Drug costs
	n	Median	Mean	SD	Median	Mean	SD	Median	Mean	SD
Mean SLEDAI score = 0	45	$7512	$14,419	$25,478	$6115	$13,047	$25,055	$773	$1372	$1503
0 < Mean SLEDAI score ≤ 2	65	$9529	$22,401	$32,079	$8211	$19,994	$30,958	$1140	$2407	$3174
2 < Mean SLEDAI score ≤ 4	34	$10,655	$22,080	$40,565	$7862	$19,721	$39,028	$1644	$2359	$2793
4 < Mean SLEDAI score	34	$13,756	$14,636	$12,963	$7616	$10,854	$12,710	$2942	$3782	$3489
Overall	178	$9655	$18,839	$29,776	$7416	$16,440	$28,850	$1526	$2399	$2927

Association between changes in SLEDAI score and changes in HRU and costs

Results from the mixed effect models showed that, with every unit decrease in the SLEDAI score, the total number of medical encounters decreases by 2.6% (p = 0.002). More specifically, every unit decrease in the SLEDAI score is associated with 6.8% (p < 0.001) fewer inpatient days, 10.0% (p < 0.001) fewer ER admissions, and 1.8% (p = 0.028) fewer outpatient visits. Based on the mixed effect models, the sample average HRU by SLEDAI scores (number of medical encounters, inpatient days, ER admissions, and outpatient visits) can be estimated by assuming all patients in the study sample had the same specific SLEDAI score. Table 5 reports the estimated sample average HRU by each SLEDAI score within the observed range of SLEDAI scores in the study sample. The table also allows for the direct estimation of the change in average HRU given a certain level of change in SLEDAI scores in the study sample. For example, a 4-point reduction from a SLEDAI score of 10 to a SLEDAI score of 6 in the study sample is associated with a decrease of 0.4 (10.0%) medical encounters over a 30-day period. An 8-point reduction from a SLEDAI score of 10 to a SLEDAI score of 2 is associated with a decrease of 0.8 (19.0%) medical encounters, and a reduction from a SLEDAI score of 10 to a SLEDAI score of 0 is associated with a decrease of 1.0 (23.3%) medical encounters. Annualized, these reductions of SLEDAI scores are associated with decreases of 5.0 medical encounters (10–6), 9.5 medical encounters (10–2), and 11.6 medical encounters (10–0), respectively (Table 5).

Table 5. Monthly HRU: Regression coefficients and average expected number of medical encounters for each SLEDAI score.

SLEDAI score	Medical encounters^a	Inpatient Days^b	ER Admissions^c	Outpatient visits^d
0	3.15	0.10	0.01	2.53
1	3.24	0.10	0.01	2.58
2	3.33	0.11	0.02	2.62
3	3.41	0.12	0.02	2.67
4	3.51	0.13	0.02	2.72
5	3.60	0.13	0.02	2.77
6	3.70	0.14	0.02	2.82
7	3.80	0.15	0.02	2.87
8	3.90	0.16	0.03	2.92
9	4.00	0.18	0.03	2.97
10	4.11	0.19	0.03	3.02
11	4.22	0.20	0.04	3.08
12	4.33	0.22	0.04	3.13
13	4.45	0.23	0.04	3.19
14	4.57	0.25	0.05	3.24
15	4.69	0.27	0.05	3.30
16	4.82	0.28	0.06	3.36
17	4.95	0.30	0.07	3.42
18	5.08	0.33	0.07	3.48

^aRegression coefficient = 0.026; p = 0.002.

^bRegression coefficient = 0.068; p < 0.001.

^cRegression coefficient = 0.100; p ≤ 0.001.

^dRegression coefficient = 0.018; p = 0.028.

Reductions in the SLEDAI score were also associated with decreases in healthcare costs. Results from the mixed effect models indicated that, with every unit reduction in the SLEDAI score, the 30-day total healthcare costs decrease by 3.8% (p = 0.005), medical costs decrease by 4.6% (p = 0.005), and drug costs decrease by 5.2% (p < 0.001). Based on the mixed effect models, the sample average healthcare costs (total costs, medical costs, and drug costs) can be estimated by assuming all patients in the study sample had the same specific SLEDAI score. Table 6 reports the estimated average healthcare costs of the sample using each SLEDAI score in the observed range of the study sample. The table also allows for the direct estimation of the change in average costs given a certain level of change the study sample SLEDAI scores in. For example, a 4-point reduction from a SLEDAI score of 10 to a SLEDAI score of 6 in the study sample is associated with a decrease of $204 (13.9%) in total healthcare costs over a 30-day period. An 8-point reduction from a SLEDAI score of 10 to a SLEDAI score of 2 is associated with a decrease of $380 (26.0%), and a 10-point reduction from a SLEDAI score of 10 to a SLEDAI score of 0 is associated with a decrease of $459 (31.3%) over a 30-day period (Table 6). Annualized, these reductions in SLEDAI scores are associated with decreases of $2485 (10–6), $4624 (10–2), and $5579 (10–0), respectively (Table 6).

Table 6. Monthly healthcare costs^a: Regression coefficients and average expected costs for each SLEDAI score.

SLEDAI score	Total costs^b	Medical costs^c	Drug costs^d
0	$1006	$780	$86
1	$1044	$817	$91
2	$1084	$855	$96
3	$1126	$896	$101
4	$1169	$938	$106
5	$1213	$982	$112
6	$1260	$1029	$118
7	$1308	$1077	$124
8	$1358	$1128	$131
9	$1410	$1182	$138
10	$1464	$1237	$146
11	$1520	$1296	$153
12	$1578	$1357	$162
13	$1639	$1421	$170
14	$1702	$1488	$180
15	$1767	$1558	$189
16	$1834	$1632	$199
17	$1905	$1709	$210
18	$1978	$1790	$222

^aCosts were measured from a payer’s perspective and reported in 2011 USD.

^bRegression coefficient = 0.038; p = 0.005.

^cRegression coefficient = 0.046; p = 0.005.

^dRegression coefficient = 0.052; p < 0.001.

Discussion

Using a unique longitudinal data source which included both HRU and cost outcomes, as well as SLEDAI scores measured at the encounter level, this is the first study to estimate the change in HRU and costs associated with a change in SLE disease activity within the same patients over time. Unlike traditional research that has established the cross-sectional correlation between disease flares or disease activity and economic outcomes across different patients, this study analyzed how HRU and cost vary over time within the same patients as their disease severity fluctuates. This study, thus, provided estimates for the amount of HRU and costs associated with decreasing disease activity.

The results suggest that potential economic savings associated with reduced disease activity could be substantial. Over a 30-day period, for a typical SLE patient, a hypothetical 4-point reduction from a SLEDAI score of 10 to a SLEDAI score of 6 is associated with a decrease of 0.4 (10.0%) in medical visits and a decrease of $204 (13.9%) in total healthcare costs. This estimate of the relationship between disease activity change within the same patients and economic outcomes is essential for the economic evaluation of medical or pharmaceutical interventions that could control or lower disease activity.

The average follow-up period was less than 3 years, and SLE severity was only measured over a 25-day period. A significant portion of the costs associated with SLE may not be observable over this short time-frame. Given the multi-organ involvement nature of SLE, healthcare costs associated with organ damage may account for a significant portion of the true costs associated with the disease. Such costs may not be related to short-term variations of disease activity, because some organ damage may be irreversible, leading to some fixed ongoing costs that might not be affected by changes in short-term SLE disease activity9. The cost changes estimated in this study may not reflect changes in organ damage status, and therefore may under-estimate the true cost savings associated with reduced SLE activity.

The observed relationship between SLE activity and costs is consistent with past cross-sectional studies. In a previous study, the authors found that the total average costs per patient-year were more than 2-fold in patients with flares than in patients without flares4. Similarly, a study presented at the 2010 European League Against Rheumatism conference used an administrative claims database to study the cost of flares of varying severity10. The average total healthcare cost per flare episode was $2163 (before adjusting for baseline factors) with mild, moderate, and severe flares costing $909, $1539, and $17,059 per episode, respectively. In addition, our calculation of average annual costs of $18,839 appears to be in line with past estimates3. We also observed that the median figure for annualized costs is associated with increased disease activity, but not mean costs. This may likely be due to the skewed costs distribution, where the mean costs can be affected by the high values of a few patients, especially among population sub-groups such as elderly patients.

This study has several limitations, such as the fact that the GHP population is from a regional health system and has limited generalizability to the entire US population. Compared to past research, the sample was relatively small and the population consisted of mostly Caucasian women who were older than the typical US SLE population and had a relatively low level of disease activity, compared to that of the Hopkins cohort7. Thus, the sample may have had limited variability in disease activity, HRU, and costs compared to more representative SLE patients. The expected total HRU and costs and their subsequent changes could be different in other samples of SLE patients, with different distributions of patient characteristics, such as age, gender, race, and comorbidities. Therefore, caution should be exercised when the study results are applied to a different sample with dissimilar patient characteristics.

Additionally, the SLEDAI score calculated for this study may be imprecise due to the retrospective study design. The SLEDAI score was originally developed for research purposes rather than for routine clinical use. SLEDAI scores were calculated based on all available recorded information, including physician notes. However, not all relevant information was recorded in the EMRs for every patient. For example, the biometrics and lab values (e.g., temperature, blood cell count, urinanalysis, CH50, and anti-DNA doubled) were often absent. This study assumed that missing lab values imply normal values, as physicians may not see the necessity of such lab test ordering with a presumed normal value. We acknowledge that our calculated SLEDAI score may under-estimate a patients’ true SLEDAI score, and may affect the estimate of the relationship between the change in the SLEDAI score and the change in healthcare costs. For example, when patients have a SLEDAI score equal to 0, they do not show recorded disease activity based on the information available in the EMR and physician notes; however, they may have residual disease activity that is not observable in this study setting. The effect of such under-estimated SLEDAI on costs is unclear. Because the SLEDAI score may be under-estimated and clustered at 0, both the absolute SLEDAI score and its economic impact may be attenuated. Further, in recent years, the SELENA SLEDAI score, a modified version of the SLEDAI score, was commonly used in research settings. It has a physical global assessment on a scale from 1–3 that has to be captured at the time of the medical encounter11. Given the retrospective nature of the study, this component could not be scored. Also, no attempt was made to retrospectively determine if a flare of SLE had occurred, which is necessary to calculate the SELENA SLEDAI scores. Therefore, the original SLEDAI score definition was used instead.

Finally, the retrospective SLEDAI score calculation may be obtained due to some subjective assessments, as raters had to determine retrospectively if some components of the SLEDAI scores were related to SLE. To assess the impact of such subjective decisions, a total of 135 encounters (∼10% of the total sample) were reviewed by at least two raters, and an analysis of inter-rater reliability was conducted. The measure of agreement between the raters varied depending on which two raters were compared. Using Fleiss’s kappa to measure overall agreement, we concluded that there was moderate agreement between the three raters. Similar results were found using Spearman correlation coefficients (results not presented). Finally, this study investigated the impact of the changes in SLEDAI score as a composite index of the level of HRU and costs, regardless of which component of the SLEDAI score fluctuated. It is possible that some SLEDAI score components have a greater impact on HRU and costs than others; these differences require further study.

Conclusion

Based on longitudinal data of SLE patients with repeated measures of disease activity over time, a lower SLEDAI score was associated with lower HRU and costs in a sample of SLE patients selected from a regional healthcare delivery system in the US.

Transparency

Declaration of funding

The funding of this research is provided by Human Genome Sciences, Inc., and GlaxoSmithKline, USA (GSK protocol number GHO-11-3363).

Declaration of financial/other relationships

HK, PJ, and CM are employees of GlaxoSmithKline, USA. SN is an employee of Human Genome Sciences, Inc.

Acknowledgments

We appreciate Saurabh Nagar of GlaxoSmithKline for programming assistance in this study.