Factors Associated With Attrition in a Longitudinal Rheumatoid Arthritis Registry

Authors


Department of Rheumatology, Brigham and Women's Hospital, 75 Francis Street, PB-B3, Boston, MA 02115. E-mail: ciannaccone@partners.org

Abstract

Objective

Loss of participants in longitudinal data collection can affect the validity of outcomes in rheumatoid arthritis (RA) registries. Prior research indicates that demographics and socioeconomic and psychosocial factors contribute to attrition. This study analyzed the characteristics of an RA registry that may contribute to attrition in a hospital-based population.

Methods

Subjects consisted of RA patients enrolled in the Brigham and Women's Rheumatoid Arthritis Sequential Study. Demographics and clinical and psychological factors were evaluated in univariate analyses to determine differences between participants who dropped out and those who completed 5 years of followup. Univariate factors with a P value <0.1 were used in a survival analysis to determine significant factors associated with attrition. A secondary analysis looked at patients who dropped out during the first year.

Results

A total of 1,144 RA participants were enrolled (509 completed 5 years of followup, 227 were still actively enrolled, and 408 dropped out). The attrition rate was 4.31% per 6-month cycle. Shorter disease duration, higher disease activity (3-variable Disease Activity Score in 28 joints using the C-reactive protein level), less education, RA drug therapy, and lower arthritis self-efficacy were statistically significant in multivariate survival analyses. In a secondary analysis, sex and age were the only additional factors found that contributed to attrition.

Conclusion

The attrition rate for this registry was similar to rates reported by other registries. Shorter disease duration, higher disease activity, and several other socioeconomic factors were associated. Men and younger patients tended to drop out during the first year. Population differences in each registry may result in different attrition patterns and ultimately, each longitudinal registry should consider conducting its own analyses.

INTRODUCTION

Observational patient registries have become a vital instrument in understanding the genetic background and treatment strategies for rheumatoid arthritis (RA) and can be valuable in researching the long-term efficacy of specific drug therapies. Multiple longitudinal RA registries have been developed in an effort to study the natural progression and treatment patterns of the disease ([1, 2]). Unlike randomized clinical trials, observational registries are considered more representative of the general population because they involve a larger group of patients. Observational registries also measure a variety of clinical outcomes, allowing researchers to study the effect of each variable on the outcome of interest ([3]).

Attrition can have a profound impact on analyses of clinical end points, making it important to understand the characteristics of patients who drop out. If patient attrition occurs randomly, it would not bias the data from the study. However, if the characteristics of the patients who drop out are systematically different from those who continue to participate, this may affect the validity of the study findings ([1]). Understanding the factors associated with attrition can aid in a more accurate interpretation of longitudinal data and help establish future preventative strategies in creating a cohort that is more representative of the population being studied.

Much of the research on attrition has been in populations involving psychological interventions or in cohorts of participants with prolonged illnesses ([4-6]). These studies suggest that study patients who drop out are less educated, may be more predisposed to depression, and may have socioeconomic and/or psychosocial obstacles that affect study participation ([4-8]). Recent studies of longitudinal RA registries and data banks report a variety of factors associated with attrition, including demographics, socioeconomic factors, and disease severity ([1, 9]). The factors associated with attrition in each study may be similar in some regard, since all include patients with RA, but there may be differences in attrition related to study protocol differences and study site location. One common theme among many attrition studies involving longitudinal registries is that the rate of attrition is higher during the beginning of the study ([1, 9]).

This study looked at the differences in demographics, disease status, comorbidities, RA treatment, and psychological and socioeconomic characteristics of RA patients enrolled in a hospital-based population over 5 years to determine factors associated with attrition. Based on previous research, our primary hypothesis is that attrition will be associated with participant demographics, RA disease activity, RA treatment, depression, and self-efficacy. In secondary analyses, we proposed examining whether the attrition rate is higher during the first year of the study by looking at the participants who drop out during the first year compared to the later dropouts and those who remained in the study for 5 years in regard to demographics, disease status, comorbidities, RA treatment, depression, and self-efficacy.

Box 1. Significance & Innovations

  • Attrition in rheumatoid arthritis (RA) registries may be affected by patient disease activity, type of RA drug therapy, and socioeconomic status.
  • Sex and age may play a role in attrition earlier on in longitudinal research registries.

PATIENTS AND METHODS

Patient recruitment

RA patients were enrolled in a large, single-center, prospective, observational patient registry in the Brigham and Women's Hospital Arthritis Center, which averages >3,700 RA patient visits per year. The registry, known as the Brigham and Women's Rheumatoid Arthritis Sequential Study (BRASS), began enrollment in March 2003. Patients (ages ≥18 years) were recruited from the practices of attending rheumatologists and fellows and screened for participation by International Classification of Diseases, Ninth Revision billing codes for diagnosis. A total of 7,898 patients with 2 separate instances of an RA billing code have been contacted by mail about enrolling in the registry. Currently, 17% of patients contacted have been enrolled in the study, 29% were excluded because they did not have a diagnosis of RA upon further medical chart review, 52% of patients were not approached for enrollment by their rheumatologist, and 2% officially refused. All diagnoses of RA were either verified according to the 1987 American College of Rheumatology criteria ([10]) by a rheumatologist or met the rheumatologists' impression for a diagnosis of RA. Information concerning patient followup has been previously reported ([3]).

Attrition

We defined attrition in our population as those who voluntarily dropped out, were lost to followup/no response, were deceased, or had a change in diagnosis of RA. Loss to followup includes patients who have moved away, those whose rheumatologist no longer provided care at the hospital, and those who never returned to be seen by a rheumatologist. If a patient does not have a return visit within 2 years, the rheumatologist is contacted and asked if the patient should be dropped from the study. Patients can drop out of the study at any time during followup.

Demographic characteristics

An annual self-administered questionnaire asked patients about their age, sex, ethnicity, education, employment, and disease duration. For the analysis, age was looked at as a continuous and categorical variable. Age, when used as a categorical variable, was grouped as ≤50, 51–60, 61–70, and ≥71 years. Education was classified into graduated high school, graduated college, and completed graduate school or higher. Employment was coded as employed, disabled, retired, and other (including housewives and students). Disease duration was also looked at as a continuous and categorical variable. Its categories were ≤5, 6–15, 16–25, and ≥26 years when used as a categorical variable.

Disease and functional status

Patients completed a blood draw at the time of the annual patient visit. This allowed the study to assess the level of disease activity by calculating the 3-variable Disease Activity Score in 28 joints using the C-reactive protein level (DAS28-CRP) ([11]). A higher DAS28-CRP score indicates worse RA disease activity. To assess functional status, the modified Health Assessment Questionnaire (M-HAQ) was also asked on the self-administered questionnaire annually ([12]). This questionnaire gives a score between 0 and 3, with a higher score indicating a worse functional status.

Forty different comorbidity diagnoses were collected by patient report annually. Some examples of the conditions that were included in our comorbidity variable are any cancer, stroke, heart disease, and diabetes mellitus. Drug information gathered at the last study visit was used to assess RA drug therapy, and medications were grouped into the following categories: tumor necrosis factor (TNF) inhibitors, biologic disease-modifying antirheumatic drugs (DMARDs), nonbiologic DMARDs, and corticosteroids.

Psychological variables

Patients answered a question concerning how often they have feelings of depression from the M-HAQ ([12]). The question reads: “We are interested in learning how your illness affects your ability to function in daily life. Please mark the response which best describes your usual abilities OVER THE PAST WEEK: Over the past week, were you able to deal with feelings of depression or feeling blue?” Answers are categorized as 1) without any difficulty, 2) with some difficulty, 3) with much difficulty, and 4) unable to do. We used this question to assess depressive mood in the cohort.

Patients also answered the other symptoms subscale of the Arthritis Self-Efficacy Scale, which assesses a patient's perceived ability to control their arthritis ([13]). This questionnaire uses a visual analog scale from 10–100, where 10 = uncertain and 100 = very certain, and asks 6 questions concerning a patient's confidence when handling certain issues involving their arthritis.

Statistical analysis

In this analysis, the last observation is the date when a patient had a 5-year followup study visit or the date when a patient dropped out. Only patients who had reached the 5-year mark were included in the 5-year completer group. Patients who were actively participating but had not reached 5 years at the time of this analysis were not included in this study. Annual attrition rates were calculated as the number of patients who dropped out in a 6-month period over person-years multiplied by 2. T-tests and chi-square tests were used to determine differences between patients who dropped out of the study and patients who completed 5 years of followup.

A multivariate survival analysis using a Cox proportional hazards survival model was chosen for this analysis. This method was selected so all available data could be used until the outcome event (dropout) was observed or reached the end of the study (5 years). In order for a variable to be included in the multivariate analysis, it had to have a P value of 0.10 or less in a univariate analysis. The multivariate survival analysis was completed using a stepwise selection of variables with a P value of 0.05 or less to remain in the model. In secondary analyses, we used a multivariate log-binomial regression model to complete a cross-sectional analysis of patients who dropped out during year 1 compared to patients who dropped out later or completed 5 years to see if the factors associated with attrition differed from the primary analysis.

RESULTS

Overall, 1,144 patients were enrolled in the registry starting in 2003. At the time of this analysis, 509 patients completed 5 years of followup, 227 patients had not completed 5 years of followup (still actively participating), and 408 patients dropped out. The mean ± SD age was 56.0 ± 14.0 years and the mean ± SD disease duration was 13.5 ± 12.3 years at baseline. Approximately 82% of patients enrolled were women and 92% were white. The mean ± SD baseline DAS28-CRP score was 3.9 ± 1.6, the mean ± SD M-HAQ score was 0.4 ± 0.5, and the mean ± SD Arthritis Self-Efficacy Scale score was 71.7 ± 19.1. All other baseline characteristics are reported in Table 1. Patients dropped out of the BRASS study for the following reasons: voluntarily dropped out (36%), lost to followup/no response (16%), died (13%), moved away (16%), rheumatologist left practice (17%), or change in diagnosis by rheumatologist (2%).

Table 1. Baseline characteristics*
 Value (n = 1,144)
  1. Values are the number (percentage) unless otherwise indicated. DAS28-CRP = 3-variable Disease Activity Score in 28 joints using the C-reactive protein level; M-HAQ = modified Health Assessment Questionnaire; ASES = Arthritis Self-Efficacy Scale.
Age, mean ± SD years56.04 ± 14.0
Age, years 
≤50379 (33.1)
51–60327 (28.6)
61–70254 (22.2)
≥71184 (16.1)
Women942 (82.3)
White1,053 (92.0)
Education 
Graduated high school254 (22.4)
Graduated college580 (51.1)
Completed graduate school301 (26.5)
Employment 
Employed530 (50.3)
Disabled101 (9.6)
Retired281 (26.7)
Other142 (13.5)
Disease duration, mean ± SD years13.54 ± 12.3
Disease duration, years 
≤5407 (35.6)
6–15321 (28.1)
16–25203 (17.8)
≥26212 (18.6)
DAS28-CRP, mean ± SD3.94 ± 1.6
M-HAQ score (range 0–3), mean ± SD0.43 ± 0.5
ASES score (range 10–100), mean ± SD71.68 ± 19.1

We compared patients enrolled from 2003–2005 with patients enrolled more recently to see if there were any differences. Patients enrolled more recently have slightly shorter disease durations, lower physician global scores, lower DAS28-CRP scores, and lower pain and M-HAQ scores. However, patients enrolled for each calendar year between 2003 and 2005 had very little difference in terms of disease duration, disease activity, and treatment.

There were also some differences in the characteristics between the patients who completed 5 years of followup and those patients who dropped out of the registry. The dropout group had a lower level of education, with 26% of the dropouts having only a high school diploma versus the 5-year completer group, which had 20% with only a high school diploma. Also, the mean ± SD disease duration was shorter in the dropout group versus the completer group (15.3 ± 13.3 years versus 19.7 ± 11.8 years). The dropout group also had worse DAS28-CRP scores, M-HAQ scores, and Arthritis Self-Efficacy Scale scores; reported more comorbidities; and was less likely to be taking a TNF inhibitor, a biologic DMARD, or methotrexate when compared to the completer group (Table 2).

Table 2. Comparison of patients who drop out versus patients who complete 5 years of followup*
 Dropouts (n = 408)5-year followup (n = 509)P
  1. DAS28-CRP = 3-variable Disease Activity Score in 28 joints using the C-reactive protein level; M-HAQ = modified Health Assessment Questionnaire; ASES = Arthritis Self-Efficacy Scale; RA = rheumatoid arthritis; TNF = tumor necrosis factor; DMARD = disease-modifying antirheumatic drug.
Age, mean ± SD years60.7 ± 16.661.0 ± 12.10.75
Women, no. (%)327 (80.2)429 (84.3)0.12
White, no. (%)368 (90.9)476 (94.3)0.06
Education, no. (%)   
High school degree105 (26.1)101 (19.9)0.04
College degree207 (51.5)264 (52.1) 
Graduate degree90 (22.4)142 (28.0) 
Employment, no. (%)   
Employed149 (39.2)227 (47.0)0.12
Disabled42 (11.1)44 (9.1) 
Retired140 (36.8)151 (31.3) 
Other49 (12.9)61 (12.6) 
Disease duration, mean ± SD years15.3 ± 13.319.7 ± 11.8< 0.0001
DAS28-CRP, mean ± SD3.64 ± 1.63.16 ± 1.4< 0.0001
M-HAQ score (range 0–3), mean ± SD0.45 ± 0.50.35 ± 0.40.0012
M-HAQ depression score (range 0–3), mean ± SD0.5 ± 0.70.37 ± 0.4< 0.003
ASES score (range 10–100), mean ± SD72.1 ± 19.476.6 ± 17.80.0003
Comorbidities, mean ± SD5.6 ± 3.44.9 ± 3.20.001
RA drug treatment, no. (%)   
TNF inhibitor115 (28)254 (50)< 0.0001
Methotrexate140 (34)267 (52)< 0.0001
Biologic DMARD123 (30)291 (57)< 0.0001
Nonbiologic DMARD229 (56)343 (67)0.0005
Corticosteroids65 (16)116 (23)0.01

Figure 1 shows the Kaplan-Meier survival curve, which displays the rate of continued participation over the 5-year period. A greater dropout rate is depicted during the first year of the study and then remains steady for the remaining 5-year period. The mean rate of attrition per 6-month followup cycle was 4.31%; however, the rate for the first 6 months was 5.4%. Of the 1,144 patients enrolled, approximately 64% either finished 5 years of followup or were still actively contributing to the registry. The rate of attrition for this RA registry was similar to rates reported in other RA registries and data banks ([1, 9]).

Figure 1.

Kaplan-Meier survival curve.

In univariate analyses for our primary hypothesis, the following factors were significantly associated with attrition: less education, higher M-HAQ depression score, shorter disease duration (categorical variable), higher DAS28-CRP score, worse M-HAQ score, lower Arthritis Self-Efficacy Scale score, higher number of comorbidities, and not taking a TNF inhibitor, methotrexate, a biologic DMARD, or a nonbiologic DMARD. Conversely, age (continuous and categorical), sex, ethnicity, employment, and not taking a corticosteroid were not significantly associated with attrition (Table 3).

Table 3. Univariate analysis*
 HR (95% CI)P
  1. HR = hazard ratio; 95% CI = 95% confidence interval; M-HAQ = modified Health Assessment Questionnaire; RA = rheumatoid arthritis; TNF = tumor necrosis factor; DMARDs = disease-modifying antirheumatic drugs; DAS28-CRP = 3-variable Disease Activity Score in 28 joints using the C-reactive protein level; ASES = Arthritis Self-Efficacy Scale.
Age, years0.99 (0.99–1.01)0.31
Female sex0.81 (0.64–1.03)0.09
Education (categorical)0.81 (0.70–0.93)0.0034
M-HAQ depression score1.28 (1.10–1.49)0.0014
Comorbidities1.05 (1.02–1.08)0.002
RA treatment  
TNF inhibitor0.50 (0.41–0.62)< 0.0001
Methotrexate0.58 (0.47–0.71)< 0.0001
Biologic DMARDs0.43 (0.35–0.53)< 0.0001
Nonbiologic DMARDs0.74 (0.61–0.90)0.002
Corticosteroids0.78 (0.60–1.02)0.07
Disease duration (categorical)0.65 (0.58–0.72)< 0.0001
DAS28-CRP1.19 (1.12–1.27)< 0.0001
M-HAQ score (range 0–3)1.43 (1.17–1.74)0.0004
ASES score (range 10–100)0.99 (0.98–0.99)0.0001

The final model for the multivariate survival analysis included the following variables: disease duration, education, TNF inhibitor, methotrexate, corticosteroid, DAS28-CRP score, and Arthritis Self-Efficacy Scale score. Patients who dropped out of the study had a shorter disease duration; had a lower level of education; were less likely to be receiving a TNF inhibitor, methotrexate, or a corticosteroid; had higher DAS28-CRP scores; and had lower Arthritis Self-Efficacy Scale scores when compared to patients who completed 5 years of followup (Table 4). Our secondary analysis included 149 patients who dropped out during the first year of followup compared to patients who dropped out later or were followed through to 5 years. The following factors were significant in the univariate analyses: younger age, higher DAS28-CRP score, shorter disease duration, less education, lower self-efficacy, higher M-HAQ score, higher M-HAQ depression score, male sex, and less TNF inhibitor, methotrexate, biologic DMARD, and corticosteroid use. In the multivariate log-binomial regression analysis, the final model showed that patients who were younger, were male, were less educated, had a shorter disease duration, did not receive TNF therapy, and had a higher DAS28-CRP score were more likely to leave the study.

Table 4. Multivariate analysis*
 HR (95% CI)P
  1. HR = hazard ratio; 95% CI = 95% confidence interval; TNF = tumor necrosis factor; ASES = Arthritis Self-Efficacy Scale; DAS28-CRP = 3-variable Disease Activity Score in 28 joints using the C-reactive protein level.
Disease duration (categorical)0.63 (0.56–0.71)< 0.0001
Education (categorical)0.85 (0.73–0.99)0.04
TNF inhibitor0.55 (0.44–0.70)< 0.0001
Methotrexate0.57 (0.46–0.71)< 0.0001
Corticosteroid0.72 (0.53–0.98)0.03
ASES score0.99 (0.99–0.99)0.02
DAS28-CRP1.21 (1.13–1.30)< 0.0001

DISCUSSION

Our attrition rates for the BRASS cohort study (4.31% per 6-month followup cycle) were similar to rates reported by other RA cohort studies, despite each study having different source populations, designs, and followup protocols ([1, 9]). For example, the Arthritis, Rheumatism, and Aging Medical Information System (ARAMIS) database reports an attrition rate of 3.8% and the National Rheumatoid Arthritis Study reports an attrition rate of 4.6% per 6-month followup cycle. Also similar to other cohort studies, the BRASS cohort attrition rate during the first year was higher than in years 2 through 5 ([1, 9]). There were some baseline differences in patients enrolling more recently compared to the patients who enrolled between 2003 and 2005. However, these differences were likely a result of changes in RA treatment that have occurred since 2003, and they did not have an effect on the overall attrition rate. In an effort to maintain patient participation in the study, we provided options to complete patient interviews on the phone, provided reimbursement for parking on the day of the clinic visit, gave small incentives and reminders of the study (pens, mugs, bags), and mailed out an annual newsletter with study updates. It is not known if these incentives played a role in patients remaining in the study.

Our primary results showed that patients who dropped out had a shorter mean disease duration compared to patients who remained in the study. However, the BRASS cohort is a registry of patients with established RA disease. Both the dropouts and the patients who completed the 5 years of study followup had lengthy disease durations, with the dropout group having a mean disease duration of 15 years compared to the patients who completed followup having a mean disease duration of 20 years. We saw similar results in our secondary analyses, which also showed that patients who dropped out in the first year had a statistically significant shorter disease duration in both the univariate and multivariate models.

In the ARAMIS database, longer disease duration was associated with decreased attrition in univariate analysis, but had no effect on the multivariate models. However, the ARAMIS group has multiple enrollment sites where the mean baseline disease duration ranges from 5 to 17 years, depending on the site ([1]). This is different from what is seen in the BRASS cohort, which has a mean baseline disease duration of 13 years. This difference might explain why shorter disease duration had a significant effect in our multivariate model, whereas in the ARAMIS study disease duration was not significant in predicting attrition.

We found that patients with worse disease activity would be more likely to drop out of the study. Higher DAS28-CRP was significant in all of our analyses, indicating that RA patients with more active disease were more likely to drop out. This could complicate long-term research looking at possible drug therapies that would improve DAS response for patients with highly active disease as well as bias the cohort, since a “healthier” population would be more likely to participate. No previous studies that we could find have looked at disease activity as a possible factor associated with attrition.

Patients who dropped out had a higher M-HAQ score in our univariate analysis only. This agrees with results found in both the ARAMIS and the National Rheumatoid Arthritis Study, which also did not find functional disability to be a predictor of attrition ([1, 9]). However, we did find that RA drug treatment is associated with attrition. Patients who took a TNF inhibitor, methotrexate, or a corticosteroid were more likely to remain in the study than the patients who dropped out. The differences in RA drug treatment may be an indication that patients who drop out could have worse physical function because their disease was being managed differently than the patients who remained in the study. Lack of TNF treatment was also significantly associated with dropping out within the first year of followup. Future research concerning drug therapy will need to take this into consideration, since the dropouts were more likely to have active disease and not be taking certain drug therapies for RA.

On average, the dropouts reported more comorbidities, but it was not statistically significant in the multivariate or regression models. No previous studies of attrition in RA have looked at comorbidities or RA drug treatment to see if there is an association with attrition. However, the overall health of a patient and the course of treatment a patient uses to treat their RA may play an integral role in attrition.

We found that participants with higher self-efficacy would be less likely to drop out of the study. This registry has a high level of reported self-efficacy, with the mean scores for the dropout group and the 5-year completer group being 72.1 and 76.6, respectively. A recent literature review by Primdahl et al looked at 74 studies reporting associations between self-efficacy and disease-related variables and found that the scores obtained by the Arthritis Self-Efficacy Scale were highly associated with disability, pain, fatigue, and disease duration ([14]). It is possible that an intervention to improve self-efficacy may improve disease variables as well, and efforts to provide additional support to patients struggling with how to deal with symptoms of their RA may be beneficial in attrition prevention.

Previous research in longitudinal psychiatric disorder studies suggested that depression is associated with attrition. In studies researching psychiatric disorders, researchers found that participants with depressive symptoms and having a major depressive disorder were more likely to be lost to followup ([5, 8]). Depressive mood based on the M-HAQ was not significantly associated with attrition in this cohort. The mean ± SD score at baseline on the M-HAQ depression question for the cohort is 0.46 ± 0.64. Based on this question, there is little depressive mood reported in the BRASS cohort. However, the M-HAQ depression question does not diagnose depression and may explain why we are not seeing a significant association with attrition. However, we did not have the ability to use confirmed diagnoses of depression in our study.

Education was also a significant factor in all of our models. The ARAMIS study and the National Rheumatoid Arthritis Study also had similar findings ([1, 9]). However, employment status was not significant except in the univariate models. It appears that only some socioeconomic variables contribute to attrition in this registry.

Only one study found that women were more likely to remain in the study ([9]). In our primary analyses, we did not see this. However, when we looked at only the dropouts from the first year, our final multivariate model included male sex as a predictor for dropping out. This may show that men are more likely to drop out of long-term studies during the first year of followup. This would present an issue because the population being studied may not be as generalizable to men, since men were more likely to drop out at the beginning of the study.

Younger age was also associated with dropping out within the first year of followup. The ARAMIS study also found that younger age was associated with attrition in their cohort ([1]). We could not confirm that age was associated overall, but it seems to be a contributing factor in patients who drop out early on in the study followup. In general, it is thought that younger patients tend to move around more, have busier schedules, and may be lost to followup more easily.

These results suggest potential issues for the BRASS cohort going forward. As we continue to enroll patients, the focus may need to be on enrolling patients who are newly diagnosed and men in order to make sure our registry is as inclusive as possible. Also, we may need to focus on preventative measures to try and decrease the overall number of patients who drop out during the first year of the study. We also need to consider our research going forward, realizing that our patient population may trend toward being more representative of patients with established, well-managed RA disease. Furthermore, our study findings suggest we need to focus on how to retain patients with heightened disease activity and lower self-efficacy. These patients may benefit from other support programs offered through the hospital and we may need to offer other services in order to make it easier for them to complete study visits.

Our study does have limitations. We have invited almost 8,000 patients to participate in the BRASS Registry, but the registry lacks information on the patients not enrolled. Selection bias of patients who did not enroll may be a concern and may lead to issues surrounding how representative the population enrolled is of all RA patients. We did not collect information on patients who were contacted but were never approached for enrollment. These patients were likely not approached for the following reasons: patients told their rheumatologist they were not interested, the rheumatologist did not feel the patient met the study eligibility criteria, or the rheumatologist was too busy to discuss the study with the patient. Specific explanations or details as to why patients dropped out of the study were also not collected. In an effort to better understand a patient's reasoning for not participating and for dropping out, collecting this information going forward is important. We also provided some incentives for patients and do not know how that affected enrollment or attrition. Finally, some variables that may have affected attrition were not included in these analyses due to missing data (e.g., income).

Similar to other studies examining attrition in RA patient populations, we found that psychosocial, socioeconomic, and demographic characteristics were associated with attrition. Our analysis also indicated that disease duration, disease activity, and differences in drug therapy were associated with attrition in this population and during the first year of followup, men and younger patients tended to drop out. This suggests that continued participation might be related to disease specific measures, whereas factors such as sex and age may play a role in the beginning of the study. Population differences in each registry may result in different attrition patterns and ultimately, each longitudinal registry should consider conducting its own analyses.

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. Ms Iannaccone had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Iannaccone, Fossel, Cui, Weinblatt, Shadick.

Acquisition of data. Iannaccone, Fossel, Cui, Weinblatt.

Analysis and interpretation of data. Iannaccone, Tsao, Cui, Weinblatt, Shadick.

ROLE OF THE STUDY SPONSOR

Crescendo Bioscience, MedImmune, and Biogen Idec had no role in the study design, data collection, data analysis, or writing of the manuscript, as well as approval of the content of the submitted manuscript. Publication of this article was not contingent on the approval of Crescendo Bioscience, MedImmune, and Biogen Idec.

Acknowledgments

We would like to thank all of our study participants.

Ancillary