Treatments and Outcomes Among Patients with Sydenham Chorea

Key Points Question Which clinical and treatment factors at onset of Sydenham chorea are associated with chorea duration, relapsing disease course, and functional outcome? Findings In this individual patient data meta-analysis of 1479 patients, those receiving at least 1 month of corticosteroids had a median chorea duration of 1.2 months vs 2.8 months for patients receiving none, a significant difference. Patients treated with antibiotics, corticosteroids, or sodium valproate had significantly reduced odds of relapse. Meaning These observational data support the use of corticosteroids, antibiotics, and sodium valproate for treatment of Sydenham chorea.


eMethods 2: Statistical Analysis
Hot-deck imputation was performed prior to multivariable modelling. 4,5Hot deck methods impute missing values within a database by using available values (donors) from the same database.No limit was set on the number of single donor selections.Case distance was weighted by univariate association of each predictor variable with the target outcome variable (time to chorea resolution at the first episode, relapsing disease course or functional outcome).Missing values were filled sequentially, starting with the variable with minimum frequency of missingness (reference imputation) (eTable 2).Outcome variables were not imputed.Following imputation, nonbinary predictor variables were recoded: worst mRS score was binarised at mRS ≥4, and age was categorised as preschool (<5 years), school age (5-11 years), adolescent (12-17 years) or adult (≥18 years), with 5-11 years as the reference category.In the Cox proportional hazard model, symptomatic medications and immunotherapy were included as time-varying features.As time-varying data (i.e.treatment start and stop times) were not always available for each treatment individually, patients treated with multiple symptomatic medications were excluded from this model, and immunotherapies (corticosteroids, IVIG and/or plasma exchange) were combined into a single time-varying feature.A single course of IVIG was regarded as having a treatment duration of one month.Time-varying features were not imputed.If a regression model failed to converge due to matrix singularity, predictor variables with zero or near-zero variance in one or both outcome classes (due to insufficient numbers of patients with the feature present) were dropped until convergence was achieved.Adjusted odds or hazard ratios were derived from the model coefficients.In the univariate survival analysis stratified by corticosteroid duration, the Tarone-Ware test was used as the primary analysis to give more weight to earlier events of chorea resolution, i.e. those which occurred during the period when patients were either receiving or not receiving steroids; the findings from this primary analysis were then validated with the log-rank test.

eMethods 3. Sensitivity Analysis
To evaluate potential biases in the regression models, secular trends were described and sensitivity analyses conducted with respect to year of disease onset and data missingness.For year of onset the dataset (cases included in ≥1 of the three multivariable models, n=618) was split by the median year of onset (2000).Missingness was compared in cases with onset before and since the year 2000 using Fisher's exact test (eTable 9) and nested regression models were created for each of the three outcomes, in which cases with pre-2000 onset were withheld (eTables 10-12).In order to test for differences in outcomes between cases with missing and complete data, the outcome distributions were compared between cases with missing data (inner distribution) and complete data (outer distribution) for each variable with missingness ≥20% in each model; variables with significant differences in the distributions (p<0.05 Variable not included in the model: immunotherapies were included as separate variables in the logistic regression models, but as a combined (time-varying) feature in the Cox proportional hazard model. 2 Variable not included in the model: no patients with benzodiazepines or antihistamines as monotherapy had data available on treatment start and stop times.

Literature Search and Record Selection: Databases Searched and Numbers of Records Retrieved, Screened and Included Database / Register Search date Search strategy Records retrieved
in Fisher's exact test for binomial outcomes or Kaplan-Meier survival analysis [log-rank test] for the eTable 1.**Papers which could be retrieved for selected languages (English, Italian, Spanish, French, Portuguese) containing individualised patient data eTable 2.

Proportions of Missing (Imputed) Data in the Multivariable Models Predictor variable Model for chorea duration at first episode Model for relapsing disease course Model for functional outcome
ARF: acute rheumatic fever; IVIG: intravenous immunoglobulin; mRS: modified Rankin Scale; SC: Sydenham's chorea.eTable 3.

THE FIRST SC EPISODE (WITHIN FIRST 3 MONTHS AFTER INITIAL PRESENTATION) Evidence of preceding Streptococcal infection
Descriptive data are provided on patients with available information, hence the varying denominators.*In 96 patients reported not to have evidence of preceding GAS infection, diagnostic criteria for ARF were met in 69/76 according to the authors.Other major features of ARF in these 96 patients included carditis/valvulitis in 36, arthritis/arthralgia in 22, and skin manifestations of ARF in three.eTable 5.

Adverse Events Associated With Immunotherapy
© 2024 Eyre M et al.JAMA Network Open.

3
Variable dropped from the model: if a model failed to converge due to matrix singularity, predictor variables with zero or near-zero variance in one or both outcome classes (due to insufficient numbers of patients with the feature present) were dropped until convergence was achieved.ARF, acute rheumatic fever; IVIG, intravenous immunoglobulin; mRS, modified Rankin Scale; SC, subcutaneous.

Nested Model for Chorea Duration at First Episode With Twentieth Century Cases Withheld
Variable dropped from the model: if a model failed to converge due to matrix singularity, predictor variables with zero or near-zero variance in one or both outcome classes (due to insufficient numbers of patients with the feature present) were dropped until convergence was achieved.

Nested Model for Relapsing Disease Course With Twentieth Century Cases Withheld
Variable dropped from the model: if a model failed to converge due to matrix singularity, predictor variables with zero or near-zero variance in one or both outcome classes (due to insufficient numbers of patients with the feature present) were dropped until convergence was achieved.It was not possible to include antibiotics in the nested model due to insufficient numbers of patients not given antibiotics in the last two decades (Figure1 *

). eTable 12. Nested Model for Poor Functional Outcome With Twentieth Century Cases Withheld
*Variable dropped from the model: if a model failed to converge due to matrix singularity, predictor variables with zero or near-zero variance in one or both outcome classes (due to insufficient numbers of patients with the feature present) were dropped until convergence was achieved.eTable 13.

Nested Model for Poor Functional Outcome With Cases Missing Data in ≥1 Missing Not at Random (MNAR) Variables Withheld
*Variable dropped from the model: if a model failed to converge due to matrix singularity, predictor variables with zero or near-zero variance in one or both outcome classes (due to insufficient numbers of patients with the feature present) were dropped until convergence was achieved.