Drug survival studies in dermatology – Principles and purposes

Research Techniques Made SimpleDrug Survival Studies in Dermatology:

Principles, Purposes and Pitfalls JMPA van den Reek, W Kievit, R. Gniadecki, J Zweegers, J.J. Goeman, PCM van de Kerkhof, MMB Seyger, EMGJ de Jong

Why Do We Need Drug Survival (DS)?

DS measures the time on-drug and covers efficacy, safety, patient and doctors’ preferences. It is a comprehensive method to evaluate treatments and differs from the classical effectiveness analyses (PASI75 in case of psoriasis) in which discontinuations, drop-outs and temporary disease-flares frequently lead to analytical problems.

When Can We Use Drug Survival?

Head-to-head comparisons between DS of different drugs (“Which drug has the longest DS?”)

Provide insight in the DS of an individual drug and identify predictors associated with DS using Cox Regression analysis (“Does weight influence drug survival?”)

Recently, DS was combined with quality of life measures. (“Do patients with long drug survival automatically have a better quality of life?”)

More possibilities need to be discovered.

What Is Drug Survival?

DS is based on regular survival analysis: a method to analyze longitudinal data for the occurrence of an ‘event’ In DS, the ‘event’ is the discontinuation of a drug If follow up of an individual patient ends early, the patient will be ‘censored’; the information until that time is fully incorporated DS analysis results in a Kaplan Meier survival curve, which can be read as illustrated in the next slide.

’

How To Read a Drug Survival Curve

What Assumptions Underlie Survival Analysis?

1.At any time point, the patients that are censored have the same survival prospects as the ones that continue.

2.Survival probabilities are the same throughout the whole study.

3.The event corresponds with the specified time and is not a raw estimation.

’Bland JM, Altman DG (1998) Survival probabilities (the Kaplan-Meier method). BMJ 317:1572.

Limitations of Drug Survival Analysis

DS can be influenced by circumstances, e.g. patient and physician’s behavior and changing circumstances in time (secular trends) The exact moment of the ‘event’ can be difficult to define or measure The predictors resulting from Cox-Regression analysis highly depend on the selection of possible predictors beforehand No predictors may be selected in a standard Cox model that have not been measured at baseline, or serious biases may occur (immortal time bias)

Ho AM,et al. (2013) Understanding immortal time bias in observational cohort studies. Anaesthesia 68:126-30.

Seven suggestions for harmonizing of drug survival studiesSuggestion Example

(1) Clearly describe: population, group characteristics, time-frame of analysis, reimbursement criteria, and existence of a drug preference policy.

(2) Diminish exclusion criteria and, instead, perform subanalyses of interest and/or multivariable regression analysis incorporating intended excluded groups.

(3) Identify possible confounders and discuss their influence.

(4) Where possible, correct for confounders by means of multivariable regression analysis.

(5) Perform sensitivity analyses where needed.

(6) Cut off analyses at the moment when a limited amount of patients becomes apparent at a specific time point.

(7) Split for reasons of discontinuation if possible.

(1) All Dutch psoriasis patients eligible for biologics using the European reimbursement criteria (ref), starting adalimumab between 2011-2013. No drug preference policy existed. Patient and treatment characteristics are shown in table x.

(2) Do not exclude biologic-naive patients beforehand, instead present all data and compare 2 groups (naive versus non-naive) or analyze influence of naivety.

(3) Discuss the possible influence that the introduction of new drugs throughout the study period has on DS.

(4) If different drugs (etanercept and adalimumab) are compared, correct for existing differences between groups (e.g. biologics naivety or age)

(5) Perform analyses with 2 different cutoff points of ustekinumab (drug with a long half-life): date of last injection versus date of last injection + t)

(6) If there are only few patients left after 5 years, censor the analysis at an earlier stage.

(7) If patients mainly discontinue due to e.g. ineffectiveness and side-effects, perform sensitivity analyses for these reasons separately.