Self-Controlled Case Series Study Methodology

The self-controlled case series (SCCS) method, often referred to simply as the case series method, is a powerful and increasingly utilized statistical methodology in pharmacoepidemiology, particularly for vaccine safety studies. This method is specifically designed to investigate the temporal association between a time-varying exposure and an acute adverse event, using data solely from individuals who have experienced the event, known as “cases”.

Key Advantages and Principles:

The SCCS method offers several distinct advantages that make it attractive for certain types of epidemiological research:

• Self-Controlled Design: A primary benefit is its self-controlled nature, meaning each individual serves as their own control. This inherently controls for all time-invariant confounders, such as sex, geographical location, genetic predispositions, and underlying health status, as these factors remain constant within an individual’s observation period.
• High Efficiency: It often boasts high efficiency when compared to traditional cohort methods, especially when the exposure risk period is short relative to the observation period.
• Reduced Data Collection: The method requires data only from cases, which significantly reduces the effort and cost associated with data collection compared to studies that require controls.
• Controlling for Time-Varying Confounders: While time-invariant confounders are implicitly controlled, time-varying confounders like age and season can be accounted for in the baseline incidence rate.

Underlying Model and Implementation:

The SCCS method is based on an underlying Poisson cohort model. For each individual in a cohort, an observation period is defined, during which event times and exposure history are recorded. This observation period is divided into age groups and risk periods. Risk periods are specific time windows during or after an exposure where an individual is hypothesized to be at an altered risk of an event, while all other times serve as control periods.

The incidence rate (λijk) is modeled multiplicatively as exp(φi + αj + βk), where:

• φi represents an effect specific to individual i.
• αj denotes an effect for age group j.
• βk signifies an effect for risk period k.

Crucially, the individual-specific effects (φi) cancel out in the conditional likelihood, demonstrating the method’s self-controlled property and its ability to adjust for time-invariant confounders. The exponentiated quantities, exp(βk), are referred to as relative incidences (RI), measuring the incidence in a risk period relative to the control period.

The multinomial likelihood can be fitted using standard statistical software packages as a Poisson regression model with a log link function. This involves formatting data with one line per interval, detailing the number of events (nijk), interval length (eijk), and factors for age and risk groups.

Key Assumptions:

For the SCCS method to be applicable, three key assumptions must be met:

1. Events arise in a non-homogeneous Poisson process: The method is suitable for independent recurrent events and can also be applied to rare, non-recurrent events. If recurrent events are not independent (e.g., multiple admissions for ITP by the same child), the analysis should ideally focus on only the first event to avoid bias.
2. The occurrence of an event must not alter the probability of subsequent exposure: This is considered the most restrictive assumption. If violated (e.g., an event prevents further treatment), modifications exist. It’s possible to test for event-dependent exposures by including a pre-exposure “risk” period and assessing if its relative incidence differs significantly from unity.
3. The occurrence of the event of interest must not censor or affect the observation period: The observation period must be independent of the event date. This assumption may be violated if the event is likely to increase the short-term death rate, though specific adaptations for such scenarios have been developed, and simulations have shown that bias can sometimes be negligible.

Comparison with Other Study Designs:

The SCCS method frequently demonstrates benefits over cohort and case-control studies:

• Cohort Studies: SCCS can achieve similar estimates to large cohort studies but with greater precision and reduced data collection effort, especially for short exposure risk periods. For instance, in a study of MMR vaccine and febrile seizures, SCCS provided a similar relative incidence but a narrower confidence interval compared to a large cohort study.
• Case-Control Studies: SCCS offers an advantage in controlling for selection and indication bias that can plague case-control studies. An example with tricyclic antidepressants and hip fracture showed a lower, potentially more accurate, relative incidence from SCCS compared to a case-control study, which was thought to be inflated by selection and indication bias. However, it is recommended to use both SCCS and cohort/case-control analyses when suitable data are available to strengthen findings or elucidate sources of bias.

The relative efficiency of the SCCS method compared to case-control and cohort methods depends on factors like the ratio of the risk period to the observation period (r), exposure prevalence (p), and the true relative incidence (eβ). SCCS is generally more efficient than the case-control method when r is small. While typically less efficient than the cohort method (except when p=1), the loss is often small when r is low and p is high.

Applications and Extensions:

Initially introduced in 1995 for vaccine adverse event studies, the SCCS method gained wider recognition through its use in studies involving the MMR vaccine and autism. Its applications have since expanded to include non-vaccine exposures, such as antidepressant use and hip fracture.

Recent developments and extensions include:

• Semiparametric approach: This avoids mis-specification of age-specific baseline incidence by leaving the age effect unspecified, providing more reliable estimates when age is a strong confounder (e.g., hepatitis B vaccination and multiple sclerosis).
• Interferent events: Adaptations for situations where an event alters subsequent exposure or observation periods, such as an intervention being contraindicated after an event or an event leading to death (e.g., Bupropion and sudden death, oral polio vaccine and intussusception).
• Sequential version (SPRT): For prospective surveillance of drug or vaccine safety, allowing for formal hypothesis testing over successive surveillance intervals (e.g., influenza vaccine and Bell’s palsy).
• Long or indefinite exposures: The method can be applied to non-acute events with long or indefinite risk periods, especially when sufficient unexposed cases are available.
• Bivariate counts: For studying the effect of exposure on two types of events simultaneously.

The SCCS method continues to evolve, with ongoing research into areas such as replacing risk periods with smooth functions, investigating the independence of multiple events, and further understanding bias from event-dependent observation periods.

Reference for the article:

Whitaker, H. J., Hocine, M. N., & Farrington, C. P. (2009). The methodology of self-controlled case series studies. Statistical Methods in Medical Research, 18(1), 7–26.