Relaxing the Rule of Ten Events per Variable in Logistic and Cox Regression

Department of Epidemiology and Biostatistics, University of California-San Francisco, 185 Berry Street, San Francisco, CA 94107, USA.
American Journal of Epidemiology (Impact Factor: 5.23). 04/2007; 165(6):710-8. DOI: 10.1093/aje/kwk052
Source: PubMed


The rule of thumb that logistic and Cox models should be used with a minimum of 10 outcome events per predictor variable (EPV),
based on two simulation studies, may be too conservative. The authors conducted a large simulation study of other influences
on confidence interval coverage, type I error, relative bias, and other model performance measures. They found a range of
circumstances in which coverage and bias were within acceptable levels despite less than 10 EPV, as well as other factors
that were as influential as or more influential than EPV. They conclude that this rule can be relaxed, in particular for sensitivity
analyses undertaken to demonstrate adequate control of confounding.

65 Reads
  • Source
    • "urbanicity, family history of mental disorders) and clinical parameters (main diagnosis, IQ, social and role functioning, duration of mental problems, type of UHR criterion, SIPS positive item scores, single APS and BIPS, SIPS subscale sum scores). For descriptive purposes, notwithstanding the recommended 5:1 relation of number of events to number of predictors (Vittinghoff and McCulloch, 2007), potential predictors showing at least a trend-level result of p b 0.10 were entered into multivariate stepwise logistic and ordinal regression analyses to detect the best, non-redundant predictor(s) of outcomes. Furthermore, the presence and size of potential age effects on outcome and the presence of BIPS and APS were examined by k × l χ 2 test and Cramer's V across 2 age groupings (AG1: 9–11, 12–14 and 15–17 years, and AG2: 9–15 and 16–17 years). "
    [Show abstract] [Hide abstract]
    ABSTRACT: Objective: The validity of current ultra-high risk (UHR) criteria is under-examined in help-seeking minors, particularly, in children below the age of 12years. Thus, the present study investigated predictors of one-year outcome in children and adolescents (CAD) with UHR status. Method: Thirty-five children and adolescents (age 9-17years) meeting UHR criteria according to the Structured Interview for Psychosis-Risk Syndromes were followed-up for 12months. Regression analyses were employed to detect baseline predictors of conversion to psychosis and of outcome of non-converters (remission and persistence of UHR versus conversion). Results: At one-year follow-up, 20% of patients had developed schizophrenia, 25.7% had remitted from their UHR status that, consequently, had persisted in 54.3%. No patient had fully remitted from mental disorders, even if UHR status was not maintained. Conversion was best predicted by any transient psychotic symptom and a disorganized communication score. No prediction model for outcome beyond conversion was identified. Conclusions: Our findings provide the first evidence for the predictive utility of UHR criteria in CAD in terms of brief intermittent psychotic symptoms (BIPS) when accompanied by signs of cognitive impairment, i.e. disorganized communication. However, because attenuated psychotic symptoms (APS) related to thought content and perception were indicative of non-conversion at 1-year follow-up, their use in early detection of psychosis in CAD needs further study. Overall, the need for more in-depth studies into developmental peculiarities in the early detection and treatment of psychoses with an onset of illness in childhood and early adolescence was further highlighted.
    Schizophrenia Research 11/2015; DOI:10.1016/j.schres.2015.10.033 · 3.92 Impact Factor
    • "The basis of such a suggestion , however, is not clear. Although simulation studies have suggested that the use of logistic regression analysis where the number of events is low might produce unstable results, this has also been disputed as a too conservative approach (Vittinghoff and McCulloch, 2007). Bigger samples and more events are almost always preferable and for this reason the results of the current study were interpreted with caution as clearly stated in the manuscript: 'Any conclusions from the current study should be viewed with caution, due to its retrospective nature. "

    Human Reproduction 10/2015; DOI:10.1093/humrep/dev262 · 4.57 Impact Factor
  • Source
    • "), even at the most optimistic scenarios, the number of events per variable should at least be greater than 5 or 10 (Vittinghoff and McCulloch, 2007) in order to reach reliable conclusions. "

    Human Reproduction 10/2015; DOI:10.1093/humrep/dev272 · 4.57 Impact Factor
Show more


65 Reads