In this free webinar, we will examine the important design considerations for analyzing recurring events and counts.
Watch the webinar at: https://www.statsols.com/en/webinar/designing-studies-with-recurrent-events
Designing studies with recurrent events (Model choices, pitfalls and group sequential design)
Students, digital devices and success - Andreas Schleicher - 27 May 2024..pptx
Designing studies with recurrent events | Model choices, pitfalls and group sequential design
1. DESIGNING STUDIES WITH RECURRENT EVENTS
Model choices, pitfalls and group sequential design
DEMONSTRATED ON
2. Head of Statistics
nQuery Lead Researcher
FDA Guest Speaker
Guest Lecturer
Webinar Host
HOSTED BY:
Ronan
Fitzpatrick
3. AGENDA
1. Recurrent Event Data and Models Introduction
2. Sample Size for Recurrent Event Models
3. Group Sequential Design for Recurrent Events
4. Conclusions and Discussion
4. The complete trial design platform to make clinical trials
faster, less costly and more successful
The solution for optimizing clinical trials
5. In 2019, 90% of organizations with clinical trials approved by
the FDA used nQuery
7. Recurrent Events Overview
Recurrent event processes are endpoints where subject can
have >1 informative event in time period
Common endpoint in clinical trials esp. chronic diseases
COPD/asthma exacerbations, MS relapses, migraines, seizures
Similar considerations and methods also relevant in case of
count data e.g. imaging, epidemiology
Should use events/counts “as-is” for better estimation but
historically often “simplified” to other endpoints
8. Recurrent Event Analysis Approaches
Event Rate Models: Estimate # of events per time unit
Parametric models for (constant over time) event rate ratio
e.g. Poisson (incl. quasi-P), negative binomial (both ZI, ZT)
Time-to-Events Models: Time(s) between/til next event
Semi-parametric models for hazard from each event
e.g. Andersen-Gill, Wei-Lin-Weissfeld, Prentice-Williams-
Peterson
Mean Cumulative Event Function: E(Events) at time t
Non-parametric method for cumulative E’s e.g. Nelson-Aalen
9. Issues in Recurrent Event Analysis
What is the question/target of interest in study?
Use event rate, time between events, # of events?
Compare groups via rate ratio, # events, intensity/HR?
Interested in first K events, later events, more intense events?
What assumptions are reasonable for data?
Independent events, non-informative censoring,
Event process differs per subject, process changes over time
Some events more important, effect of terminal events
10. Count Model Examples
1 Poisson Model Event Rates, rate ratio, constant rate
2 Negative Binomial Event Rates, RR, event rate/subject
3 Andersen-Gill Time-to-Events, intensity/hazard of all E
4 Wei-Lin-Weissfeld Time-to-Events, time til 1st K events, HR
5 Prentice-Williams-Peterson Time-to-Events, gap between 1st K E’s, HR
6 MCF Non-parametric, # events at time t
7 Approximate Models Survival models (1st E), t-test, event “rates”
12. Example Models for SSD
1) Poisson: Signorini (1991), Lehr (1992), Gu et .al. (2008)
Parametric model for rate ratio, assumes constant rate ratio
Rate constant over time, same rate per subject unless quasi-P
2) Negative Binomial: Zhu & Lakkis (2014), Tang (2015,18)
Parametric model for rate ratio, assumes constant rate ratio
Rate constant over T, Poisson rate/subject from gamma dist.
3) Andersen-Gill: Bernardo & Harrington (2001), Tang &
Fitzpatrick (2019)
Semi-Parametric for intensity/hazard, assume constant RR here
Arbitrary rate over T (e.g. Weibull), need proportional hazards
13. Sample Size for Recurrent Events
SSD initially focussed on simple
approx. assuming Poisson (e.g. √)
More work done for Poisson
regression (& Quasi-Poisson
Recent work on Neg. Binomial and
extend it & Poisson for unequal
follow-up, dropout & NI/equiv
SSD also derived for Andersen-Gill
incl. flexible approach for design
and hazard function assumptions
Source: R. Lehr
(1992)
Source: H. Zhu & H. Lakkis
(2014)
Source: Y. Tang
(2015)
Source: D. Signorini
(1991)
Source: Y. Tang & R. Fitzpatrick
(2019)
14. Worked Example 1
“On the basis of previous studies of
fluticasone propionate–salmeterol
combinations we assumed a yearly
exacerbation rate with vilanterol of 1·4
and a dispersion parameter of 0·7. Thus,
we calculated that a sample size of 390
assessable patients per group in each
study would provide each study with 90%
power to detect a 25% reduction in
exacerbations in the fluticasone furoate
and vilanterol groups versus the vilanterol
only group at a two-sided 5% significance
level”
Source: Lancet Respiratory Medicine
(2013)
Parameter Value
Significance Level (Two-Sided) 0.05
Control Incidence Rate (per year) 1.4
Rate Ratio 0.75
Exposure Time (Years) 1
Dispersion Parameter 0.7
Power (%) 90%
Negative Binomial/ Recurrent Events Example
16. Group Sequential Designs (GSD)
“Group sequential trials can provide
ethical and efficiency advantages by
reducing the expected sample size and
calendar time of clinical trials and by
accelerating the approval of effective
new treatments.
For example, a group sequential design
with a single interim analysis and can
reduce the expected sample size of the
trial by roughly 15 percent relative to a
comparable non-adaptive trial.”
GSD allows interim analyses to stop
early
Interim: look while trial is on-going
Looks occur at pre-specified times
E.g. After 50% subjects measured
Can stop for benefit and/or futility
Neither? Continue until end/next look
However, need to account for effect of
multiple analyses
Do by “spending” error at each look
17. GSD Details
2 early stopping criteria (work similar)
1. Efficacy (α-spending)
2. Futility (β-spending)
Use “spending function” to account for
multiple analyses/”chances”
Spend proportion of total error at each look
More conservative N vs. fixed term analysis
Multiple SF available (incl. custom)
O’Brien-Fleming (conservative), Pocock
(liberal), Power, Hwang-Shih-DeCani (flexible)
Different heuristics for efficacy/futility
Conservative SF expected for efficacy
More flexibility for futility (less problematic)
𝛼 𝜏 = 2 1 − Φ
𝑧 𝛼/2
𝜏
O’Brien-Fleming Spending Function
18. Recurrent Events GSD Issues
Appropriate MLE and test for GST
Normal approx. if t same per subject?
For NB, no closed form if t differs
How turn info. time into “real” time
Info. time related to follow-up times,
n, rates and dispersion parameter
Equal t ~ Means, Unequal ~ TTE
Type I Error ↑ if low N/high κ
No Exact, some proposed altered stats
t-distribution GSD better for low N
Source: Mütze, T., Glimm, E., Schmidli, H., &
Friede, T. (2019).
19. Group Sequential Design for Negative Binomial
Parameter Value
Number of Looks 3
Efficacy Bound O’Brien Fleming
Futility Bound Non-Binding
Beta Spending Function Hwang-Shih-DeCani
HSD Parameter -1.25
Extend previous example to group
sequential design with 2 interim analyses
with O’Brien Fleming efficacy bound and
non-binding Hwang-Shih-DeCani futility
bound with gamma = -1.25
Worked Example 2
Source: Lancet Respiratory Medicine
(2013)
Parameter Value
Significance Level (Two-Sided) 0.05
Control Incidence Rate (per year) 1.4
Rate Ratio 0.75
Exposure Time (Years) 1
Dispersion Parameter 0.7
Power (%) 90%
20. Discussion and Conclusions
Recurrent Events/Count data common in clinical trials
Move away from approximations to modelling process fully
Variety of models depending on scientific question
Parametric: event rate; Semi-parametric: t between/to events
SSD methods available for most common models
Negative Binomial, Poisson, Andersen-Gill; one less barrier
GSD methods available for recurrent events growing
Approx. work adequately for standard Phase III assumptions
24. The solution for optimizing clinical trials
PRE-CLINICAL
/ RESEARCH
EARLY PHASE
CONFIRMATORY
POSTMARKETING
Animal Studies
ANOVA / ANCOVA
1000+ Scenarios for Fixed Term,
Adaptive & Bayesian Methods
Survival, Means, Proportions &
Count endpoints
Sample Size Re-Estimation
Group Sequential Trials
Bayesian Assurance
Cross over & personalized medicine
CRM
MCP-Mod
Simon’s Two Stage
Cohort Study
Case-control Study
25. Cameron, A. C., & Trivedi, P. K. (2013). Regression analysis of count data. Cambridge
university press.
Cook, R. J., & Lawless, J. (2007). The statistical analysis of recurrent events. Springer Science
& Business Media.
The Analysis of Recurrent Events: A Summary of Methodology; J Rogers, Oxford:
https://www.psiweb.org/docs/default-source/resources/psi-subgroups/scientific/2016/time-
to-event-and-recurrent-event-endpoints/jrogers.pdf?sfvrsn=a86d2db_2
Lehr, R. (1992). Sixteen S‐squared over D‐squared: A relation for crude sample size estimates.
Statistics in medicine, 11(8), 1099-1102.
Signorini, D. F. (1991). Sample size for Poisson regression. Biometrika, 78(2), 446-450.
Gu, K., Ng, H. K. T., Tang, M. L., & Schucany, W. R. (2008). Testing the ratio of two poisson
rates. Biometrical Journal, 50(2), 283-298.
References
26. References
Zhu, H. (2017). Sample size calculation for comparing two poisson or negative binomial rates
in noninferiority or equivalence trials. Statistics in Biopharmaceutical Research, 9(1), 107-
115.
Zhu, H., & Lakkis, H. (2015). Sample size calculation for comparing two negative binomial
rates. Statistics in medicine, 33(3), 376-387.
Tang, Y. (2015). Sample size estimation for negative binomial regression comparing rates of
recurrent events with unequal follow-up time. Journal of biopharmaceutical statistics, 25(5),
1100-1113.
Tang, Y. (2017). Sample size for comparing negative binomial rates in noninferiority and
equivalence trials with unequal follow-up times. Journal of biopharmaceutical statistics, 1-17.
Tang, Y., & Fitzpatrick, R. (2019). Sample size calculation for the Andersen‐Gill model
comparing rates of recurrent events. Statistics in Medicine, 38(24), 4819-4827.
27. References
Dransfield, M. T., et. al. (2013). Once-daily inhaled fluticasone furoate and vilanterol versus
vilanterol only for prevention of exacerbations of COPD: two replicate double-blind, parallel-
group, randomised controlled trials. The lancet Respiratory medicine, 1(3), 210-223.
Jennison, C., & Turnbull, B. W. (1999). Group sequential methods with applications to clinical
trials. CRC Press.
Cook R. & Lawless J. (1996). Interim monitoring of longitudinal comparative studies with
recurrent event responses. Biometrics 52: 1311–1323.
Mütze, T., Glimm, E., Schmidli, H., & Friede, T. (2019). Group sequential designs for negative
binomial outcomes. Statistical methods in medical research, 28(8), 2326-2347.
Mutze, T., E. Glimm, H. Schmidli, and T. Friede. (2019). Group sequential designs with robust
semiparametric recurrent event models. Statistical Methods in Medical Research