Mortality outcomes in pediatric rheumatology in ... - Wiley Online Library

ARTHRITIS & RHEUMATISM Vol. 62, No. 2, February 2010, pp 599–608 DOI 10.1002/art.27218 © 2010, American College of Rheumatology

Mortality Outcomes in Pediatric Rheumatology in the US Philip J. Hashkes,1 Bridget M. Wright,2 Michael S. Lauer,3 Sarah E. Worley,1 Anne S. Tang,1 Philip A. Roettcher,4 and Suzanne L. Bowyer5 23 (21%), and were unknown in 12 patients (11%). Rheumatic diagnoses, age at diagnosis, sex, and early use of systemic steroids and methotrexate were significantly associated with the risk of death. Conclusion. Our findings indicate that the overall mortality rate for pediatric rheumatic diseases was not increased. Even for the diseases and conditions associated with increased mortality, mortality rates were significantly lower than those reported in previous studies.

Objective. To describe mortality rates, causes of death, and potential mortality risk factors in pediatric rheumatic diseases in the US. Methods. We used the Indianapolis Pediatric Rheumatology Disease Registry, which includes 49,023 patients from 62 centers who were newly diagnosed between 1992 and 2001. Identifiers were matched with the Social Security Death Index censored for March 2005. Deaths were confirmed by death certificates, referring physicians, and medical records. Causes of death were derived by chart review or from the death certificate. Standardized mortality ratios (SMRs) and 95% confidence intervals (95% CIs) were determined. Results. After excluding patients with malignancy, 110 deaths among 48,885 patients (0.23%) were confirmed. Patients had been followed up for a mean ⴞ SD of 7.9 ⴞ 2.7 years. The SMR of the entire cohort was significantly decreased (0.65 [95% CI 0.53–0.78]), with differences in patients followed up for >9 years. The SMR was significantly greater for systemic lupus erythematosus (3.06 [95% CI 1.78–4.90]) and dermatomyositis (2.64 [95% CI 0.86–6.17]) but not for systemic juvenile rheumatoid arthritis (1.8 [95% CI 0.66–3.92]). The SMR was significantly decreased in pain syndromes (0.41 [95% CI 0.21–0.72]). Causes of death were related to the rheumatic diagnosis (including complications) in 39 patients (35%), treatment complications in 11 (10%), non-natural causes in 25 (23%), background disease in

The practice of pediatric rheumatology includes more than 170 conditions, both inflammatory and noninflammatory (1). Approximately 3 in 1,000 children have a rheumatic condition (1). While we tend to study pediatric rheumatology outcomes in terms of remission versus active disease, organ and radiologic damage, function, and quality of life, there is a small but significant increase in mortality rates among these patients. An increased rate of mortality has been found in juvenile rheumatoid arthritis (JRA) (2–12), childhood systemic lupus erythematosus (SLE) (13–21), dermatomyositis (DM) (22–25), various vasculitides (26–30), and systemic sclerosis (SSc) (31–33). However, most of those studies were of relatively small cohorts, reported mortality outcomes only on specific diseases, had a followup time of ⬍10 years, and were conducted prior to the 1990s, when new therapies were developed. Even larger studies were flawed; most were based on physician surveys and questionnaires with no strategies to verify response accuracy (4,8). The diagnoses in studies of national cohorts were usually not assigned by pediatric rheumatologists (9,11). There are no data on the mortality rate from many rare rheumatic inflammatory diseases (primary vasculitis and autoinflammatory diseases) and noninflammatory conditions, including pain syndromes (such as fibromyalgia). The causes of death were usually not adequately verified, and no systematic attempt was made to look for potential

Supported by the Northeast Ohio Chapter of the Arthritis Foundation. 1 Philip J. Hashkes, MD, MSc, Sarah E. Worley, MS, Anne S. Tang, MS: Cleveland Clinic, Cleveland, Ohio; 2Bridget M. Wright, MD: The Children’s Hospital at The Medical Center of Central Georgia, Macon; 3Michael S. Lauer, MD: National Heart, Lung and Blood Institute, Bethesda, Maryland; 4Philip A. Roettcher, MSOR, MBA: Indiana University, Indianapolis; 5Suzanne L. Bowyer, MD: James W. Riley Hospital, Indianapolis, Indiana. Address correspondence and reprint requests to Philip J. Hashkes, MD, MSc, Consultant, Department of Rheumatic Diseases A50, Cleveland Clinic Foundation, 9500 Euclid Avenue, Cleveland, OH 44195. E-mail: [email protected]. Submitted for publication March 25, 2009; accepted in revised form October 16, 2009. 599

600

HASHKES ET AL

risk factors or predictors of mortality early in the disease course. Therefore, we performed a systematic study of mortality outcomes for all pediatric rheumatic conditions in the US based on the world’s largest pediatric rheumatology registry. Our specific aims were to estimate the mortality rate of patients with pediatric rheumatic conditions, to describe the causes of death, and to search for possible mortality risk factors early in the disease course. PATIENTS AND METHODS Indianapolis Pediatric Rheumatology Disease Registry (PRDR) data. We used the PRDR, which prospectively collected data on new patients from 62 pediatric rheumatology centers in the US between February 1992 and January 2001. Data were collected on 49,023 patients at their first clinic visit, with no followup data (34). We excluded 138 patients with a diagnosis of malignancy, and included the remaining 48,885 patients in the analysis. Identification of deceased patients. Patients in the PRDR were identified only by their initials and birth date. The full identity was known only to the referring pediatric rheumatologist/center. Due to 2003 Health Insurance Portability and Accountability Act (HIPAA) rules prohibiting transferring data on living patients for research purposes without specific consent (not obtained for the PRDR, which was established in 1992), we performed the search without identifying living patients. We matched the database list of initials and birth dates to a list of all deaths in the US Social Security Death Index (SSDI) corresponding to the range of birth dates in the PRDR database. The SSDI contains the records of deceased persons who possessed Social Security numbers and whose death had been reported to the Social Security Administration. Figure 1 shows a flow chart of the process of determining the number of deceased patients. The death certificates obtained from the National Death Index of the National Center for Health Statistics were reviewed independently by 3 physicians (PJH, BMW, and SLB), who were blinded with regard to the original diagnosis in the database, using standard rules for association to a diagnosis of a rheumatic disease. These rules were based on the cause of death and the proximity of the state where the death occurred to the center where the patient was seen. We identified deaths that were definitely, probably, and possibly associated with rheumatic conditions. We then reviewed the PRDR entry for these deceased patients to match the sex and diagnosis to the death certificate and contacted referring physicians for confirmation. In order to detect deceased patients who may have been missed in our search, we performed a sensitivity analysis by asking all referring center physicians whether they were aware of any deaths among former patients. We then checked whether they had been reported to the PRDR. Only 5 deceased patients were identified by this method, and 4 of these were from 1 center. The medical charts of 87 deceased patients (79%) were reviewed to confirm the

Figure 1. Flow chart showing the process of detection of deceased patients from the Indianapolis Pediatric Rheumatology Disease Registry (PRDR). Definite, probable, and possible refer to deaths that were definitely, probably, and possibly associated with rheumatic conditions.

initial diagnosis, cause of death, and demographic, clinical, and treatment factors related to these patients. Identification of causes of death. The cause of death was obtained from the patient’s chart and/or autopsy, when available. For other patients we used the cause stated in the death certificate. The cause of death was independently classified by 4 physicians (PJH, BMW, MSL, and SLB) into causes related to the rheumatic diagnosis (including disease complications), treatment complications, non-natural causes (e.g., accidents, homicide, and suicide), background disease, and unknown/unclear causes. The certainty of the cause of death was classified into definite (for example, autopsy-proven, microbiology-proven, or non-natural deaths), probable (clinical circumstantial evidence), possible (lack of evidence other than death certificate), and unknown. Identification of potential risk factors/predictors of mortality. The PRDR included data for almost all patients on demographic (age, sex, and ethnicity) and diagnostic factors (rheumatic diagnoses based on International Classification of Diseases, Ninth Revision codes; time of disease onset; and diagnosis). There were additional queries that changed every 1–2 years (34). We obtained data on patient zip code and medical insurance as surrogates for socioeconomic data. (Other surrogates were missing from almost all charts.) Statistical analysis. Expected survival for the entire cohort, for specific disease categories, specific diseases, and

MORTALITY OUTCOMES IN PEDIATRIC RHEUMATOLOGY IN THE US

inflammatory versus noninflammatory conditions was computed from the age- and sex-matched US population and compared with observed survival using the 1-sample log rank test (35,36); for nearly 10% of the patients information on ethnicity was missing, so this factor was not used in this computation. The standardized mortality ratio (SMR) comparing observed to expected survival was computed, with 95% confidence intervals (95% CIs) calculated from the Poisson CI on the number of deaths. The association between survival times and potential risk factors was assessed using a separate Cox proportional hazards model for each risk factor, with time to death as the dependent variable, censored at March 1, 2005 (6 months prior to the SSDI search date of September 1, 2005), since the SSDI is sensitive in detecting deaths up to 6 months prior to this date (37,38). Time 0 was defined as the date of clinic visit rather than the date of onset of disease, since the date of onset is often a subjective judgment and was not available for many patients. To assess predictors of survival multivariably while adjusting for and assessing the potential correlation between the outcomes of subjects treated at the same center, we used the robust sandwich estimate for the covariance matrix (39). Only variables for data that were missing on ⬍5% of observations and, in the case of categorical variables, had ⱖ5% of observations in each level of the variable were used in the modeling process. The appropriateness of the proportional hazards assumption was assessed graphically using log–log survival plots and by entering risk factor–by-time interactions into the model, and model fit was assessed using analysis of residuals. The final model was limited to 11 variables (10% of 110 events) to avoid overparameterization. To improve the robustness of our model, parameter estimates to the effects of influential observations, the final model was fit using 1,000 samples bootstrapped from the data with replacement, and the mean, 2.5th percentile, and 97.5th percentile of the parameter estimates from those models were used to obtain final esti-

601

mates and 95% CIs for the hazard ratios (HRs) of the risk factors. For review of the medical charts of the deceased patients, we used descriptive statistics. We obtained median household incomes for US census data for deceased patients’ zip codes of residence and compared the distribution of these median incomes to 2006 household income quintiles for the US using a chi-square goodness-of-fit test (http://www.census. gov/hhes/www/income/histinc/h01AR.html). All analyses used complete cases, and all tests were 2-tailed and performed at a significance level of 0.05. SAS, version 9.2 (SAS Institute, Cary, NC) and R 2.3.1 (R Foundation for Statistical Computing, Vienna, Austria) were used for analysis. Oracle and SQL software were used for the SSDI and PRDR list match.

RESULTS Summary data from the PRDR. Data were collected on 48,885 patients (34). There were 56,260 rheumatic diagnoses in 48,322 patients, with 6,947 patients having more than one diagnosis. Of the 39,221 patients for whom information was available, 24,911 (63.5%) had chronic inflammatory diseases. The most common inflammatory diagnoses were JRA (in 9,894 of the 24,911 patients with a diagnosis of inflammatory disease [39.7%]), SLE (in 1,440 [5.8%]), Raynaud’s phenomenon (in 1,222 [4.9%]), Henoch-Scho ¨nlein purpura (HSP; in 849 [3.4%]), juvenile DM (in 686 [2.8%]), and scleroderma (in 419 [1.7%]). The most common noninflammatory diagnoses included arthralgia (in 5,112 of the 14,310 patients with noninflammatory disease [35.7%]), antinuclear antibody positivity (in 4,346 [30.4%]), hyper-

Table 1. Summary statistics from the Indianapolis Pediatric Rheumatology Disease Registry cohort* Entire cohort Age at visit, years Age at death, years Time from onset to diagnosis, months Weight, kg Height, cm Time to travel to medical center, minutes† Distance from medical center, miles† Sex, no. (%) female/male Ethnicity, no. (%) Caucasian African American Hispanic Asian Native American Other/mixed

10.0 ⫾ 4.7 (10.3 [0–28]); 48,207 – 12.6 ⫾ 20.3 (5.0 [0–243.2]); 27,352 38.8 ⫾ 21.4 (35.0 [1–179]); 6,794 133.7 ⫾ 27.1 (136.0 [52.5–200]); 2,865 64.1 ⫾ 64.1 (45.0 [0–1,440]); 6,618 81.6 ⫾ 480.2 (25.0 [0–10,000]); 641 30,865 (64.2)/17,181 (35.8) 35,032 (79.5) 5,121 (11.6) 2,297 (5.2) 732 (1.7) 59 (0.1) 831 (1.9)

Deceased patients

P

11.6 ⫾ 5.1 (12.7 [0.2–21.6]); 110 16.6 ⫾ 6.1 (17.4 [0.9–30.4]); 110 10.6 ⫾ 14.3 (4.5 [1.0–76.1]); 56 39.2 ⫾ 19.7 (42.0 [12.0–71.0]); 9 160.3 ⫾ 11.3 (160.3 [152.3–168.3]); 2 73.3 ⫾ 92.4 (20 [20–180]); 3 40.0 ⫾ 10.0 (40 [30–50]); 3 62 (56.4)/48 (43.6)

0.001 – 0.29 0.95 0.18 0.88 0.037 0.084 0.17

75 (71.4) 21 (20) 7 (6.7) 2 (1.9) 0 (0) 0 (0)

* Except where indicated otherwise, values are the mean ⫾ SD (median [range]); number of patients. The earliest date of birth of patients in the registry was February 16, 1970, and the latest date of birth was May 31, 2001 (for 48,427 patients). The earliest date of visit was January 11, 1990, and the latest date of visit was November 2, 2001 (for 48,721 patients). The earliest date of death was March 2, 1992, and the latest date of death was February 18, 2005 (for 110 patients). Ethnicity was unknown for 5 of the 110 deceased patients. † Some of the patients traveled from outside the US.

602

HASHKES ET AL

Table 2. SMRs for the entire cohort and for diagnosis subgroups* Observed deaths, no. (%)

Expected deaths

SMR (95% CI)

P

110 (0.23) 17 (1.2) 5 (0.8) 19 (0.2) 7 (0.2) 6 (0.6) 12 (0.1) 5 (0.2) 4 (0.1) 4 (1.9) 5 (0.2) 3 (0.1) 3 (1.4) 6 (0.1) 6 (0.3) 24 (0.5) 14 (0.4) 64 (0.3) 29 (0.2)

169 5.6 1.9 33.5 16.3 3.3 29.2 8.5 17.7 0.8 10.2 10.4 0.9 13.1 8.7 19.7 13.5 84.4 50.4

0.65 (0.53–0.78) 3.06 (1.78–4.90) 2.64 (0.86–6.17) 0.57 (0.34–0.89) 0.43 (0.17–0.88) 1.8 (0.66–3.92) 0.41 (0.21–0.72) 0.59 (0.19–1.38) 0.23 (0.06–0.58) 4.71 (1.28–12.07) 0.49 (0.16–1.14) 0.29 (0.06–0.84) 3.47 (0.72–10.15) 0.46 (0.17–0.99) 0.69 (0.25–1.51) 1.22 (0.78–1.81) 1.04 (0.57–1.74) 0.76 (0.58–0.97) 0.58 (0.39–0.83)

⬍0.001 ⬍0.001 0.030 0.014 0.025 0.15 0.002 0.24 0.003 0.002 0.11 0.031 0.031 0.055 0.37 0.34 0.88 0.027 0.003

Entire cohort (n ⫽ 47,449) SLE (n ⫽ 1,393) Dermatomyositis (n ⫽ 662) JRA (all) (n ⫽ 9,604) Other arthritis (n ⫽ 4,614) Systemic JRA (n ⫽ 962) Pain syndromes (all) (n ⫽ 8,147) Fibromyalgia (n ⫽ 2,297) Arthralgia (n ⫽ 5,324) Primary vasculitis (n ⫽ 206)† Orthopedic/mechanical disorders (n ⫽ 2,401) Infection (n ⫽ 2,857) Genetic/chromosomal/metabolic/bone dysplasia (n ⫽ 208) Laboratory abnormalities (n ⫽ 4,466) Other rheumatic diagnosis (n ⫽ 2,286) Other nonrheumatic diagnosis (n ⫽ 5,316) Diagnosis not clear/unspecified (n ⫽ 3,829) Inflammatory disease (n ⫽ 24,187) Noninflammatory disease (n ⫽ 13,920)

* Only patients with complete data on visit date, date of birth, and sex were included, so the numbers of patients do not equal the overall number of patients in the Indianapolis Pediatric Rheumatology Disease Registry. This table includes data primarily on disease categories. For specific diseases/disease subtypes we included those we thought would be of particular interest. For other diseases/subtypes the standardized mortality ratio (SMR) was not statistically significant. The SMR is defined as the number of observed deaths divided by the number of expected deaths. 95% CI ⫽ 95% confidence interval; SLE ⫽ systemic lupus erythematosus; JRA ⫽ juvenile rheumatoid arthritis. † Except for Kawasaki disease and Henoch-Scho ¨nlein purpura.

Figure 2. Kaplan-Meier survival curves for the entire cohort (n ⫽ 47,449) (A), for patients with systemic lupus erythematosus (n ⫽ 1,393) (B), for patients with systemic juvenile rheumatoid arthritis (n ⫽ 962) (C), and for patients with primary vasculitis other than Kawasaki disease or Henoch-Scho ¨nlein purpura (n ⫽ 206) (D). These plots include only patients for whom complete data on date of visit, date of birth, and sex were available, so the numbers of patients do not equal the overall numbers in the Indianapolis Pediatric Rheumatology Disease Registry.

mobility syndrome (in 2,416 [16.9%]), and fibromyalgia (in 2,234 [15.6%]).

The SSDI search was performed a mean of 7.9 ⫾ 2.7 years after the patients were registered (median 7.8


603

Table 3. Causes of death by diagnostic category* Cause of death

Diagnosis

Related to diagnosis assigned by a rheumatologist

Rheumatology treatment–related

Non-natural

Background (or other) disease

JRA and other chronic arthritis (n ⫽ 20) Connective tissue diseases (n ⫽ 26) Vasculitis (n ⫽ 5) Infection (n ⫽ 3) Orthopedic/mechanical/bone dysplasia (n ⫽ 2) Pain syndromes (n ⫽ 17) Laboratory abnormalities (n ⫽ 2) Other rheumatic diagnosis (n ⫽ 2) Other nonrheumatic diagnosis (n ⫽ 14)

7 (35) 15 (58) 4 (80) 0 (0) 1 (50) 0 (0) 0 (0) 0 (0) 10 (71)

4 (20) 5 (19) 1 (20) 1 (33) 0 (0) 0 (0) 0 (0) 0 (0) 0 (0)

5 (25) 2 (8) 0 (0) 2 (67) 1 (50) 10 (59) 1 (50) 1 (50) 0 (0)

4 (20) 4 (15) 0 (0) 0 (0) 0 (0) 7 (41) 1 (50) 1 (50) 4 (29)

* Values are the number (%) of patients. Numbers of patients differ slightly from those shown in other tables due to differences between the Indianapolis Pediatric Rheumatology Disease Registry diagnosis (which includes ⬎1 diagnosis for some deceased patients) and the diagnosis obtained from the chart review. Only the primary diagnosis was used in this analysis. JRA ⫽ juvenile rheumatoid arthritis.

[range 0–14.6 years]), with no significant differences between the various diagnostic categories. Other summary data are detailed in Table 1. Mortality rates. We identified 110 deceased subjects (0.23% [95% CI 0.19–0.27]). The mortality rate of the PRDR cohort was significantly lower than the expected mortality rate from the age- and sex-adjusted US population (Table 2 and Figure 2A). However, significant differences in observed and expected mortality rates were seen only among the 18,111 patients who were followed up for at least 9 years. For several diagnostic categories and specific diseases, we found a significant change in the population SMR (Table 2 and Figures 2B–D). Among disease categories significant increases in the population SMR and in mortality rates were found for connective tissue diseases and primary vasculitis (except for Kawasaki disease and HSP) when compared with the rest of the PRDR cohort. The SMR and the mortality rate were significantly decreased for pain syndromes and arthralgia compared with the rest of the PRDR cohort. Among specific diseases the SMR was significantly increased for SLE and DM. For other specific diseases, including all subtypes of JRA, Kawasaki disease, and HSP, the SMR and the mortality rate did not significantly differ from that of the rest of the PRDR cohort. Sixty-four (58%) of the deaths were in patients with a primary diagnosis of inflammatory disease, and 45 (41%) were in patients with a primary diagnosis of noninflammatory disease. (The primary diagnosis was unknown in 1 patient.) The SMRs of both the noninflammatory and inflammatory groups were significantly lower than the general population (but the difference was more significant for the noninflammatory group).

The 5-year survival rate was 99.9% (95% CI 99.8–99.9) for the entire cohort, 99.5% (95% CI 99.1– 99.9) for SLE, 99.6% (95% CI 99.4–99.9) for all connective tissue diseases, 99.8% (95% CI 99.5–100) for systemic JRA, and 99.0% (95% CI 97.7–100) for primary vasculitis other than Kawasaki disease and HSP (Figures 2A–D). The 10-year survival rate was 99.7% (95% CI 99.6–99.8) for the entire cohort, 98.2% (95% CI 97.2– 99.2) for SLE, 98.8% (95% CI 98.2–99.3) for all connective tissue diseases, 99.1% (95% CI 98.3–99.8) for systemic JRA, and 96.7% (95% CI 93.2–100) for primary vasculitis other than Kawasaki disease and HSP. The mean time to death for the entire cohort was 5.0 ⫾ 3.4 years (median 4.9 [range 0–12.2 years]). No significant differences were seen between those with inflammatory conditions and those with noninflammatory conditions. Causes of death. The causes of death were related to the diagnosis for which the deceased was seen by a rheumatologist (including complications) in 39 patients (35%), treatment complications in 11 (10%), non-natural causes in 25 (23%), background disease in 23 (21%), and were unknown/unclear in 12 (11%). The cause of death was ascertained from the chart in only 29 patients (26%); among these 11 (38%) were confirmed by an autopsy. The cause of death for the other 81 patients (74%) was obtained from the death certificate, except for 3 patients, for whom no cause of death was available. The cause of death was definitive for 47 patients (43%), probable for 31 patients (28%), possible for 23 patients (21%), and unknown or unclear for 9 patients (8%). There was a significant correlation between the rheumatic diagnosis and the cause of death (Table 3).

604

Among the 64 deceased patients in whom inflammatory disease was diagnosed, 28 (44%) died of a cause related to the diagnosis for which they were seen by a rheumatologist, while only 11 (24%) of the 45 patients in whom a noninflammatory disease was diagnosed died of a cause directly related to the diagnosis. In contrast, 13 (29%) and 14 (31%), respectively, of the patients in whom a noninflammatory disease was diagnosed died of a non-natural cause or of a background (or other) disease, as opposed to 11 (17%) and 9 (14%), respectively, among patients in whom an inflammatory disease was diagnosed (P ⬍ 0.001). This was particularly evident for patients diagnosed with a pain syndrome, among whom 59% died of non-natural causes and 41% of background (or other) disease. Among patients with SLE, 6 died of renal disease, 2 of pancreatitis, 1 of pulmonary hemorrhage, and 1 of intracranial hemorrhage. Four patients with SLE died of infection; of these, 2 patients had had bone marrow transplantation. Two patients with systemic JRA died of macrophage activation syndrome, 2 of heart disease, 1 of infection, and 1 of a secondary malignancy. Two patients with DM died of heart disease, 1 of a myocardial infarction, and 1 of myocarditis. One died of aspiration pneumonia, 1 of gastrointestinal perforation, and 1 of a surgical complication related to aortic coarctation (DiGeorge syndrome). Six patients with pain syndromes committed suicide, and 2 died of homicide. Among the patients with vasculitis who died, most causes of death were related to their disease, but causes were well described for only 2 patients. One patient with polyarteritis nodosa died of a myocardial infarction, and a patient with HSP died of gastrointestinal perforation and renal failure. Predictors of mortality. When comparing characteristics of deceased patients with survivors in the PRDR, only the age at visit and distance from the center were significant factors. An older age at the time of visit and less distance from the pediatric rheumatology center were associated with increased mortality (Table 1). The other factors that we examined, including sex, ethnicity, time from onset of disease to diagnosis, weight, height, initial medication use, and time required to get to the treating center were not significantly associated with mortality risk. In univariable survival analysis, several diagnostic categories and diseases were significant predictors of a higher risk of mortality. These included all connective tissue diseases (HR 4.5 [95% CI 2.9–6.9]), SLE (HR 6.0 [95% CI 3.6–10.1]), DM (HR 3.3 [95% CI 1.3–8.0]), primary vasculitis, except for Kawasaki disease and HSP

HASHKES ET AL

Table 4. Significant HRs for diagnoses and other predictors of mortality versus survival in a multivariable survival model* Variable

HR (95% CI)

Older age at visit (⬎14.5 years) Arthralgia Pain syndromes Connective tissue disease Infection Juvenile rheumatoid arthritis Other arthritis Other nonrheumatic diagnosis Unspecified diagnosis Male sex

2.3 (1.6–3.3) 0.43 (0.09–0.97) 0.83 (0.37–1.5) 4.8 (2.8–8.4) 0.55 (0.00–1.5) 1.2 (0.62–2.2) 0.83 (0.25–1.8) 2.5 (1.4–4.0) 2.0 (0.92–3.9) 1.7 (1.1–2.5)

* HRs ⫽ hazard ratios; 95% CI ⫽ 95% confidence interval.

(HR 9.2 [95% CI 3.4–25.0]), systemic JRA (HR 2.5 [95% CI 1.1–5.7]), and genetic/chromosomal/metabolic diseases (HR 6.2 [95% CI 2.0–19.5]). Diagnoses of arthralgia (HR 0.3 [95% CI 0.1–0.9]) were significantly predictive of better survival; pain syndromes were marginally predictive (HR 0.6 [95% CI 0.3–1.1]). Other significant predictors of mortality included an older age at the time of the first visit to a rheumatologist (HR 1.1 [95% CI 1.0–1.1] per year of age) and the use of systemic steroids (HR 5.6 [95% CI 1.1–28.9]) or methotrexate (HR 15.1 [95% CI 2.9–77.7]) at the initial visit. Other disease categories and specific diseases, sex, ethnicity, height, weight, year of visit, distance and time from rheumatology center, time from diagnosis, erythrocyte sedimentation rate, and presence of antinuclear antibody were not associated with increased mortality risk. The number of patients was too small for calculation of HRs for other factors in the PRDR. In the multivariable model only connective tissue diseases, other nonrheumatic diagnosis, male sex (even though not significant in the univariable analysis), and older age at visit were significantly associated with mortality, while arthralgia was a negative predictor of mortality (Table 4). Review of the charts of deceased patients. We reviewed charts for 87 patients (79%) (5 only partially). Seventeen charts (15%) were not located or had been destroyed. For 3 patients there were no pediatric rheumatologists currently in that locality, and for 3 patients there was lack of cooperation or institutional review board refusal to allow chart review. The deceased patients had resided in 30 states, with the mode from Ohio (17 patients [15%]). These numbers correlated with the number of patients submitted to the PRDR from each state (data not shown). Thirty (27%) of the deceased patients were between ages 17 and 21 years (“transition”


years); of these 15 (50%) died of disease complications or infections, and 10 (33%) died of non-natural causes. Among the 71 deceased patients for whom data were available, 58 (82%) lived in urban localities, and 13 (18%) lived in rural localities. Of the 76 deceased patients for whom zip code of residence was known (69%), 2 (3%) lived in zip codes with median household income in the lowest socioeconomic quintile, 36 (47%) lived in zip codes with median income in the second quintile, 32 (42%) lived in zip codes with median income in the middle quintile, 6 (8%) lived in zip codes with median income in the fourth quintile, and none lived in zip codes with median income in the highest quintile. These patients were significantly more likely to reside in zip codes with middle-quintile median household incomes than the general population (P ⬍ 0.001). We obtained insurance information on 79 (72%) of the deceased patients; 53 (67%) had commercial insurance/pay-for-service and 26 (33%) had Medicaid (or lacked insurance). The proportion of deceased patients who had been on Medicaid was similar to that reported for the general population for the study period (http://www.cbpp.org/8-29-06health.htm). Data on clinical and laboratory parameters and treatments are available from the corresponding author upon request. DISCUSSION This is the largest systematic mortality outcome study in pediatric rheumatology published to date. Overall, we did not detect an increase in mortality compared with the general population, and even for those diseases and conditions associated with increased mortality the rates were significantly lower than those reported in previous studies, especially for systemic JRA, SLE, DM, and vasculitis. Previous studies of JRA showed a decreasing mortality rate between the early 1970s and early 1990s, but still demonstrated a significantly increased SMR (7–11). In the literature on childhood SLE, 5-year survival rates range from 83% to 95%, and 10-year survival rates range from 76% to 95% (13–21). The most recent study of juvenile DM demonstrated a mortality rate of ⬃1.5–2.5% over 3 to 5 years (25). Mortality rates were very low for HSP and Kawasaki disease (28,29) but very high for polyarteritis nodosa (30). For SSc the 5-year survival rate is ⬃90%, the 10-year survival rate is 80%, and the 20-year survival rate is 69% (31–33). One possible cause of the increased survival in the present study compared with previous studies may be the improved treatment that was introduced in the

605

1990s. Alternative reasons may be the relatively short mean followup time in the cohort examined in the present study and the underestimation of mortality due to methodology limitations (discussed below). Patients with arthralgia in the present study had a decreased mortality rate. The reason for this is not clear, but many of these patients see numerous physicians (including gastroenterologists, neurologists, psychologists, and psychiatrists), and perhaps increased medical vigilance was related to the decreased mortality rate (40). Adult studies have shown an increased mortality in pain syndromes (41). As expected, most of the patients with inflammatory disease died of their disease or disease complications. With longer followup this proportion may change, given the possibility of secondary malignancies or an increased rate of infections related to prolonged immunosuppression. Interestingly, most of the patients who had pain syndromes died of non-natural causes. Many children with pain syndromes have mood disorders, such as anxiety and/or depression, that may result in an increased incidence of death from non-natural causes, such as suicide, homicide, and even motor vehicle accidents resulting from suicidal intent or substance abuse. We were unable to obtain data from these patients’ charts to support this hypothesis. Unlike in earlier studies (2,3,6,9), no patients with JRA in the present study died of amyloidosis, late development of other autoimmune disease, or suicide (9,10). Two patients with systemic JRA died of macrophage activation syndrome, which is probably currently the most common cause of death in patients with systemic JRA (12). With regard to early factors associated with mortality, we found an increased risk of mortality among male patients and among patients first seen at an older age. These factors remained significant even after adjustment for diagnosis in a multivariable model. We do not have an explanation for these findings, although population mortality rates in children are slightly greater in males than in females. An interesting finding was the increased mortality in patients living closer to the medical center, although data were available for a small fraction of the PRDR cohort (insufficient for inclusion in the predictor model). One possible explanation may be the existence of many tertiary hospitals (where most pediatric rheumatologists practice) in inner city neighborhoods. Alternatively, being able to travel from a great distance to a tertiary medical center may be a surrogate indicator of higher socioeconomic status, as was found in other mortality studies (42). According to a recent pediatric

606

rheumatology workforce analysis, the mean distance traveled by patients to a pediatric rheumatologist is 57 miles (43). In our limited data we did not find that socioeconomic factors (median income in the zip code of residence and proportion of patients on Medicaid) played a major role in mortality. This study has limitations that may have resulted in underestimation of mortality. External validity was shown, however, by the survival of patients in our cohort who were eventually diagnosed with malignancies, which were not included in the primary analysis since they are not considered pediatric rheumatic conditions, being within the range of modern childhood malignancy studies (44,45). Of 138 patients with malignancy, 17 died. The 5-year survival rate was 90.1% (95% CI 85.1–95.4), and the 10-year survival rate was 85.2% (95% CI 78.6– 92.3). Face validity was also evident from the finding that patients with inflammatory conditions had a higher mortality rate than those with pain syndromes. Our ability to detect deceased patients was decreased due to the limited identifiers of patients in the PRDR. Directly approaching physicians who had contributed to the PRDR for their list of deceased patients would have been less accurate (including physician memory bias) and may have led to compliance issues (e.g., retired physicians, patients that had moved, locating patients on physician master lists, and time constraints). We could not ask physicians to identify their complete list of patients due to HIPAA regulations that prohibit research on living patients without specific consent. A recent publication of the Institute of Medicine pointed out the difficulties HIPAA rules impose on research (46). Identifying the deceased via the SSDI also has limitations, since patients without a Social Security number were missed. These may include some foreignborn patients (such as illegal immigrants), Native Americans (a small number in our cohort), and patients whose guardians or relatives had not requested death benefits or who had changed their name/initials (due to marriage). Despite these limitations, several studies have shown the SSDI to be 86–97.5% sensitive and 99% specific when searched for information on deaths that occurred up to 6 months prior to the search date (37,38,47). In addition, a Social Security number is obtained for nearly all children in the US in the first year of life (since it is needed for child exemptions on tax forms), and most families utilize Social Security death benefits. Pediatric rheumatic diseases and mortality are very rare in the first year of life (1). We also performed

HASHKES ET AL

a sensitivity analysis (as described in Patients and Methods) for patients who were not detected in the search results and found only 5 additional deceased patients. Since, as our study demonstrated, mortality is relatively uncommon in the individual pediatric rheumatology practice, it is expected that most physicians would remember their deceased patients from the last 10–12 years. There may be problems in generalizing our data. Many US centers did not participate in the PRDR, and the compliance of participating centers in submitting patients varied considerably. However, the PRDR, which includes ⬃49,000 patients, is the world’s largest registry for these conditions. Since the PRDR mainly includes patients first seen in pediatric rheumatology clinics, some rheumatic diseases, especially those with a hospital-based presentation (such as SLE) or those seen frequently by other pediatric specialists (such as Kawasaki disease, rheumatic fever, and HSP) may be underrepresented if they were not followed up in the clinic or registered during clinic followup. Since this study was not a population study, it is possible that cases with mild disease that were not referred to pediatric rheumatologists were missed. This referral bias may offset other factors that may have resulted in an underestimation of mortality. The mortality rate and causes of death may not be generalized to other parts of the world. Studies have shown higher rates of JRA-associated amyloidosis in Europe and suicide among patients with JRA in Finland (2,3,9). Risk factors may differ between countries. However, differences in areas such as approach to treatment have diminished recently with improved international information sharing. The earliest date of death in our cohort was 1992, when the incidence of amyloidosis was already decreasing in Europe and international cooperation was increasing. Future studies should compare our results with results from existing registries in other countries (48,49). Due to the relatively short followup period, we did not capture the full extent of mortality, especially late deaths after early adulthood. Thus, we may have missed the mortality risk of premature cardiovascular diseases known to occur in SLE and RA. A potential strength of the present study is the accuracy of the diagnoses, since patients were diagnosed by qualified pediatric rheumatologists. However, patients’ diagnoses may have changed since the first visit. Indeed, in 36 (41%) of the 87 deceased patients whose charts were reviewed in detail the diagnosis was changed. Also, for ⬃15% of the patients in the PRDR,


there was no definitive diagnosis. This limited our ability to use database diagnoses for analysis. Unfortunately, we were not able to sample the charts of the surviving patients for diagnostic accuracy, as previously described (46). Sources of the cause of death are often inaccurate, including death certificates, which often report ill-defined causes of death. In only a minority of cases we confirmed the cause from patient charts and/or autopsy reports. However, we were able to report a high degree of certainty of the cause of death for 71% of the patients. Since the information in the PRDR was limited, we could not explore in depth for risk factors or early predictors of mortality. For many factors data were available for only a minority of patients. The chart review was also limited by lack of complete data for each patient. Other factors should be investigated in future registries or in case–control studies. While the results of our study are encouraging, with the mortality rate of our entire cohort similar to that of the age- and sex-matched US population, it is important to follow up this cohort in the future for mortality trends, especially later deaths seen as sequelae in many rheumatic diseases.

607

Medical School, Durham, NC), Thomas J. A. Lehman, MD (Hospital for Special Surgery, New York, NY), Calvin B. Williams, MD, PhD (Milwaukee Children’s Hospital, Milwaukee, WI), Beth S. Gottlieb, MD, MS (Schneider Children’s Hospital, New Hyde Park, NY), Deborah Rothman, MD, PhD (Shriner’s Hospital, Springfield, MA), David M. Siegel, MD, MPH (Strong Memorial Hospital, University of Rochester, Rochester, NY), Paula W. Morris (Arkansas Children’s Hospital, Little Rock, AR), Leonard D. Stein, MD (University of North Carolina, Chapel Hill), Donald P. Goldsmith, MD (St. Christopher’s Hospital, Philadelphia, PA), Linda WagnerWeiner, MD (La Rabida Hospital, University of Chicago, Chicago, IL), Richard K. Vehe, MD (Gillette Children’s Hospital, University of Minnesota, St. Paul, MN), Kathleen M. O’Neil, MD (Children’s Hospital of Buffalo, Buffalo, NY), Lawrence S. Zemel, MD (Connecticut Children’s Medical Center, Hartford, CT). We also thank Ms Christine Skibinski who formed and helped operate the database. AUTHOR CONTRIBUTIONS All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Hashkes and Ms Worley had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study conception and design. Hashkes, Wright, Lauer, Worley. Acquisition of data. Hashkes, Wright, Roettcher, Bowyer. Analysis and interpretation of data. Hashkes, Lauer, Worley, Tang.

REFERENCES ACKNOWLEDGMENTS We wish to thank the following investigators of the PRDR, who assisted us in identifying the deceased and helped us in the chart review process, including obtaining approval by Institutional Review Boards, hosting the authors during the chart review visit, and reviewing several of the charts: Hermine Brunner, MD, MSc (Cincinnati Children’s Hospital Medical Center, Cincinnati, OH), Gloria C. Higgins, MD, PhD (Children’s Hospital of Columbus, Columbus, OH), Larry B. Vogler, MD (Egleston Children’s Hospital at Emory University [ARPOC], Atlanta, GA); Carol A. Wallace, MD (Seattle Children’s Hospital Center, Seattle, WA), Jorge LopezBenitez, MD (Floating Children’s Hospital, Boston, MA), Donna L. Gibbas, MD (ARPRC, Atlanta, GA), Carol B. Lindsley, MD (University of Kansas Medical Center, Kansas City, KS), J. Kenneth Herd, MD (East Tennessee State, Johnson City, TN), Marisa S. Klein-Gitelman, MD, MPH (Children’s Memorial Hospital, Chicago, IL), Terry L. Moore, MD (St. Louis Medical Center, St. Louis, MO), Linda K. Myers, MD (University of Tennessee, Memphis, TN), Harry L. Gewanter, MD and Eugenio Monasterio, MD (Children’s Hospital of Richmond, Richmond, VA), Kenneth N. Schikler, MD (Kosair Children’s Hospital, Louisville, KY), David Sherry, MD and Terri H. Finkel, MD, PhD (Children’s Hospital of Philadelphia, Philadelphia, PA), Andreas A. Reiff, MD (Children’s Hospital of Los Angeles, Los Angeles, CA), Ilona S. Szer, MD (Children’s Hospital of San Diego, San Diego, CA), Egla C. Rabinovich, MD, MPH (Duke University

1. Petty RE, Cassidy JT. Introduction to the study of rheumatic diseases in children. In: Cassidy JT, Petty RE, editors. Textbook of pediatric rheumatology. 4th ed. Philadelphia: WB Saunders; 2001. p. 4–8. 2. Laaksonen AL. A prognostic study of juvenile rheumatoid arthritis: analysis of 544 cases. Acta Paediatr Scand 1966;Suppl 166: 1–163. 3. Ansell BM, Wood PH. Prognosis in juvenile chronic polyarthritis. Clin Rheum Dis 1976;2:397–412. 4. Baum J, Gutowska G. Death in juvenile rheumatoid arthritis. Arthritis Rheum 1977;20 Suppl:253–5. 5. Bernstein B. Death in juvenile rheumatoid arthritis. Arthritis Rheum 1977;20 Suppl:256–7. 6. Bywaters EG. Death in juvenile chronic polyarthritis. Arthritis Rheum 1977;20 Suppl:256. 7. Stoeber E. Prognosis in juvenile rheumatoid arthritis: follow-up of 433 chronic rheumatoid children. Eur J Pediatr 1981;135:225–8. 8. Wallace CA, Levinson JE. Juvenile rheumatoid arthritis: outcome and treatment for the 1990s. Rheum Dis Clin North Am 1991;17: 891–905. 9. Savolainen HA, Isomaki HA. Decrease in the number of deaths from secondary amyloidosis in patients with juvenile rheumatoid arthritis. J Rheumatol 1993;20:1201–3. 10. French AR, Mason T, Nelson AM, O’Fallon WM, Gabriel SE. Increased mortality in adults with a history of juvenile rheumatoid arthritis: a population-based study. Arthritis Rheum 2001;44: 523–7. 11. Thomas E, Symmonds DP, Brewster DH, Black RJ, Macfarland GJ. National study of cause-specific mortality in rheumatoid arthritis, juvenile chronic arthritis, and other rheumatic conditions: a 20 year followup study. J Rheumatol 2003;30:958–65. 12. Stephan JL, Kone-Paut I, Galambrun C, Mouy R, Bader-Meunier

608

13. 14. 15. 16. 17.

18. 19. 20.

21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31.

HASHKES ET AL

B, Prieur AM. Reactive hemophagocytic syndrome in children with inflammatory disorders: a retrospective study of 24 patients. Rheumatology (Oxford) 2001;40:1285–92. Meislin AG, Rothfield NF. Systemic lupus erythematosus in childhood: analysis of 42 cases with comparative data on 200 adult cases followed concurrently. Pediatrics 1968;42:37–49. Walravens PA, Chase HP. The prognosis of childhood systemic lupus erythematosus. Am J Dis Child 1976;130:929–33. Kaslow RA, Masi AT. Age, sex, and race effects on mortality from systemic lupus erythematosus in the United States. Arthritis Rheum 1978;21:473–9. Abeles M, Urman JD, Weinstein A, Lowenstein M, Rothfield NF. Systemic lupus erythematosus in the younger patients: survival studies. J Rheumatol 1980;7:515–22. Glidden RS, Mantzouranis EC, Borel Y. Systemic lupus erythematosus in childhood: clinical manifestations and improved survival in fifty-five patients. Clin Immunol Immunopathol 1983;29: 196–210. Lehman TJ, McCurdy DK, Bernstein BH, King KK, Hanson V. Systemic lupus erythematosus in the first decade of life. Pediatrics 1989;83:235–9. Lacks S, White P. Morbidity associated with childhood systemic lupus erythematosus. J Rheumatol 1990;17:941–5. Tucker LB, Menon S, Schaller JG, Isenberg DA. Adult and childhood onset systemic lupus erythematosus: a comparison of onset, clinical features, serology and outcome. Br J Rheumatol 1995;34:866–72. Ward MM, Pyun E, Studenski S. Causes of death in systemic lupus erythematosus: long-term followup of an inception cohort. Arthritis Rheum 1995;38:1492–99. Bitnum S, Daeschner CW Jr, Travis LB, Dodge WF, Hopps HC. Dermatomyositis. J Pediatr 1964;64:101–31. Bowyer SL, Blane CE, Sullivan DB, Cassidy JT. Childhood dermatomyositis: factors predicting functional outcome and development of dystrophic calcinosis. J Pediatr 1983;103:882–8. Hochberg MC. Epidemiology of polymyositis/dermatomyositis. Mt Sinai J Med 1988;55:447–52. Pachman LM. Juvenile dermatomyositis: immunogenetics, pathophysiology, and disease expression. Rheum Dis Clin North Am 2002;28:579–602. Counahan R, Winterborn MH, White RH, Heaton JM, Meadow SR, Bluett NH, et al. Prognosis of Henoch-Schonlein nephritis in children. Br Med J 1977;2:11–4. Goldstein AR, White RH, Akuse R, Chantler C. Long-term follow-up of childhood Henoch-Schonlein nephritis. Lancet 1992; 339:280–2. Stewart M, Savage JM, Bell B, McCord B. Long term renal prognosis of Henoch-Schonlein purpura in an unselected childhood population. Eur J Pediatr 1988;147:113–5. Nakamura Y, Yanagawa H, Harada K, Kato H, Kawasaki T. Mortality among persons with a history of Kawasaki disease in Japan: the fifth look. Arch Pediatr Adolesc Med 2002;156:162–5. Besbas N, Ozen S, Saatci U, Topaloglu R, Tinaztepe K, Bakkaloglu A. Renal involvement in polyarteritis nodosa: evaluation of 26 Turkish children. Pediatr Nephrol 2000;14:325–7. Foeldvari I, Zhavania M, Birdi N, Cuttica RJ, de Oliveira SH, Dent PB, et al. Favorable outcome in 135 children with systemic sclerosis: results of a multi-national survey. Rheumatology (Oxford) 2000;39:556–9.

32. Scalapino K, Arkachaisri T, Lucas M, Fertig N, Helfrich DJ, Londino AV Jr, et al. Childhood onset systemic sclerosis: classification, clinical and serologic features, and survival in comparison with adult onset disease. J Rheumatol 2006;33:1004–13. 33. Martini G, Vittadello F, Kasapcopur O, Magni Manzoni S, Corona F, Duarte-Salazar C, et al. Factors affecting survival in juvenile systemic sclerosis. Rheumatology (Oxford) 2009;48:119–22. 34. Bowyer S, Roettcher P, for the Pediatric Rheumatology Database Research Group. Pediatric rheumatology clinic populations in the United States: results of a 3 year survey. J Rheumatol 1996;23: 1968–74. 35. Therneau T, Offord J. Expected survival based on hazard rates (update): technical report series no. 63. Rochester (MN): Department of Health Sciences Research, Division of Biostatistics, Mayo Clinic; 1999. 36. Arias E, Anderson RN, Kung HC, Murphy SL, Kochanek KD. Deaths: final data for 2001. Natl Vital Stat Rep 2003;52:1–115. 37. Hauser TH, Ho KK. Accuracy of on-line databases in determining vital status. J Clin Epidemiol 2001;54:1267–70. 38. Schisterman EF, Whitcomb BW. Use of the Social Security Administration Death Master File for ascertainment of mortality status. Popul Health Metr 2004;2:2. 39. Lin DY, Wei LJ. The robust inference for the proportional hazards model. J Am Stat Assoc 1989;84:1074–8. 40. Wolfe F, Anderson J, Harkness D, Bennett RM, Caro XJ, Goldenberg DL, et al. A prospective, longitudinal, multicenter study of service utilization and costs in fibromyalgia. Arthritis Rheum 1997;40:1560–70. 41. Macfarlane GL, McBeth J, Silman AJ. Widespread body pain and mortality: prospective population based study. BMJ 2001;323: 662–5. 42. Shishehbor MH, Litaker D, Pothier CE, Lauer MS. Association of socioeconomic status with functional capacity, heart rate recovery, and all-cause mortality. JAMA 2006;295:784–92. 43. Deal CL, Hooker R, Harrington T, Birnbaum N, Hogan P, Bouchery E, et al. The United States rheumatology workforce: supply and demand, 2005–2025. Arthritis Rheum 2007;56:722–9. 44. Linabery AM, Ross JA. Childhood and adolescent cancer survival in the US by race and ethnicity for the diagnostic period 19751999. Cancer 2008;113:2575–96. 45. Pulte D, Gondos A, Brenner H. Trends in 5- and 10-year survival after diagnosis with childhood hematologic malignancies in the United States, 1990-2004. J Natl Cancer Inst 2008;100:1301–9. 46. Nass SJ, Levit LA, Gostin LO, editors. Effect of the HIPAA privacy rule on health research. In: Beyond the HIPAA privacy rule: enhancing privacy, improving health through research. Washington (DC): The National Academies Press; 2009. p. 199–244. 47. Lash TL, Silliman RA. A comparison of the National Death Index and Social Security Administration databases to ascertain vital status. Epidemiology 2001;12:259–61. 48. Malleson PN, Fung MY, Rosenberg AM. The incidence of pediatric rheumatic disease: results from the Canadian Pediatric Rheumatology Association Disease Registry. J Rheumatol 1996; 23:1981–7. 49. Symmons DP, Jones M, Osborne J, Sills J, Southwood TR, Woo P. Pediatric rheumatology in the United Kingdom: data from the British Pediatric Rheumatology Group National Diagnostic Register. J Rheumatol 1996;23:1975–80.