Relaxing the Rule of Ten Events per Variable in Logistic and Cox Regression
The DOI matches its Crossref record; 9 of the 9 references it cites were found in a registry or an index.
Abstract
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.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“Following the generally accepted rule of thumb that 5–10 events are required per covariate in logit or Cox regression models (Peduzzi et al., 1995; Peduzzi et al., 1996; Vittinghoff & McCulloch, 2007), we assessed whether or not each paper could have supported fully saturated regression models with product terms between all of the psychosocial problems considered.”
From Syndemics of psychosocial problems and HIV risk: A systematic review of empirical tests of the disease interaction concept (2015) · cites this paper as Vittinghoff & McCulloch, 2007
In the citing paper
“Because so few students used psychotherapy, we limited the number of explanatory variables to the number of observed events divided by five (14).”
From Mental health treatment seeking among veteran and civilian community college students (2017) · cites this paper as [14]
In the citing paper
“This sample size was chosen taking into account approximate sample size estimation criteria [38,39]. We worked with the criteria that 10 subjects would be needed per parameter included in a statistical model used to adjust for confounding.”
From Relationships, love and sexuality: what the Filipino teens think and feel (2009) · cites this paper as [38]
In the citing paper
“As Vittinghoff and McCulloch 27 recommended, the 10 outcome events per predictor variable rule was relaxed.”
From Mental health services for parents who lost a child to cancer: if we build them, Will they come? (2015) · cites this paper as [27]
In the citing paper
“The requirements regarding events per variable were fulfilled owing to a considerable number of injuries in the present dataset.34”
From How much running is too much? Identifying high-risk running sessions in a 5200-person cohort study (2025) · cites this paper as [34]
Citations
Cites
A simulation study of the number of events per variable in logistic regression analysis
Peduzzi, P., Concato, J., Kemper, E., Holford, T. R., & Feinstein, A. R. (1996) · Journal of Clinical Epidemiology · doi:10.1016/s0895-4356(96)00236-3 · Cited by 9,655 · cited 1×
“(2, 3) more closely, we also examined models with all binary predictors.”
BackgroundModeling and variable selection in epidemiologic analysis
Greenland, S. (1989) · American Journal of Public Health · doi:10.2105/ajph.79.3.340 · Cited by 2,208
Peduzzi, P., Concato, J., Feinstein, A. R., & Holford, T. R. (1995) · Journal of Clinical Epidemiology · doi:10.1016/0895-4356(95)00048-8 · Cited by 2,154 · cited 2×
“(2, 3) more closely, we also examined models with all binary predictors.”
BackgroundResultThe Abuse of Power: The Pervasive Fallacy of Power Calculations for Data Analysis
Hoenig, J. M., & Heisey, D. M. (2001) · The American Statistician · doi:10.1198/000313001300339897 · Cited by 2,046
Sample-Size Formula for the Proportional-Hazards Regression Model
Schoenfeld, D. A. (1983) · Biometrics · doi:10.2307/2531021 · Cited by 1,096
Concato, J., Peduzzi, P., Holford, T. R., & Feinstein, A. R. (1995) · Journal of Clinical Epidemiology · doi:10.1016/0895-4356(95)00510-2 · Cited by 797 · cited 1×
“The rule of thumb that logistic and Cox models should be used with a minimum of 10 events per predictor variable (EPV) is based on two simulation studies (1–3).”
MethodSection 4 Long-term survival results in medically and surgically randomized patients
Detre, K. M., Peduzzi, P. N., Scott, S. M., & Davies, B. (1985) · Progress in Cardiovascular Diseases · doi:10.1016/0033-0620(85)90019-2 · Cited by 24
Section 2 Design and baseline characteristics
Peduzzi, P. N., Detre, K. M., & Gage, A. A. (1985) · Progress in Cardiovascular Diseases · doi:10.1016/0033-0620(85)90017-9 · Cited by 15
Harrell, F., Jr. (1996) · doi:10.1002/(sici)1097-0258(19960229)15:4<361::aid-sim168>3.0.co;2-4
Cited by
Items 201–300 of 1,018
Analysis of Risk Factors for Glans Dehiscence After Tubularized Incised Plate Hypospadias Repair
SNODGRASS, W. T., Cost, N. G., Nakonezny, P. A., & Bush, N. C. (2011) · The Journal of Urology · doi:10.1016/j.juro.2010.12.070 · Cited by 95
Mental illness stigma, secrecy and suicidal ideation
Oexle, N., Ajdacic-Gross, V., Kilian, R., Müller, M., Rodgers, S., Xu, Z., Rössler, W., & Rüsch, N. (2017) · Epidemiology and Psychiatric Sciences · doi:10.1017/s2045796015001018 · Cited by 93
“However, while earlier work recommended the 10 : 1 rule (ten events per parameter; Peduzzi et al. 1996), as illustrated by Vittinghoff & McCulloch (2007) a 5 : 1 rule is legitimate.”
BackgroundAcute Compartment Syndrome in Children and Teenagers With Tibial Shaft Fractures
Shore, B. J., Glotzbecker, M. P., Zurakowski, D., Gelbard, E., Hedequist, D., & Matheney, T. H. (2013) · Journal of Orthopaedic Trauma · doi:10.1097/bot.0b013e31828f949c · Cited by 93
“We limited the number of candidate variables in the multivariable analysis to ensure reliable estimation of the regression coefficients.(31) Therefore, 4 covariates of interest were tested (those that were significant or approaching significance in the univariate analysis: age, physis status (open vs.”
MethodPecorelli, N., Hershorn, O., Baldini, G., Fiore, J. F., Stein, B., Liberman, A. S., Charlebois, P., Carli, F., & Feldman, L. S. (2016) · Surgical Endoscopy · doi:10.1007/s00464-016-5169-2 · Cited by 103
“According to previous research, between 5 and 10 events per independent variable are required to obtain reliable regression coefficients in multivariate logistic regression models [22], and a minimum of 15 subjects per variable are required for linear regression [23].”
MethodMcCREDIE, V. A., Ferguson, N. D., Pinto, R. L., Adhikari, N. K. J., Fowler, R., Chapman, M., Burrell, A., BAKER, A. J., Cook, D. J., Meade, M. O., et al. (2017) · Annals of the American Thoracic Society · doi:10.1513/annalsats.201608-620oc · Cited by 103
Deutsch, T., Lippmann, S., Frese, T., & Sandholzer, H. (2015) · Scandinavian Journal of Primary Health Care · doi:10.3109/02813432.2015.1020661 · Cited by 92
Rodgers, R. F., Ziff, S., Lowy, A. S., Yu, K., & Austin, S. B. (2017) · International Journal of Eating Disorders · doi:10.1002/eat.22682 · Cited by 100
Systematic review of prediction models for gestational hypertension and preeclampsia
Antwi, E., Amoakoh‐Coleman, M., Vieira, D. L., Madhavaram, S., Koram, K. A., Grobbee, D. E., Agyepong, I. A., & Klipstein‐Grobusch, K. (2020) · PLoS ONE · doi:10.1371/journal.pone.0230955 · Cited by 114
“To prevent overestimation of risks by prediction models, it is recommended that the number of outcomes in relation to the number of predictors (events-per-variable) should be at least ten to one [69,70].”
BackgroundJärvinen, T., Ilonen, I., Kauppi, J. T., Salo, J., & Räsänen, J. V. (2018) · World Journal of Surgical Oncology · doi:10.1186/s12957-018-1327-4 · Cited by 92
“75 events per variable (47 events, 4 variables), which is over the suggested limit of 10 (Table 4) [27].”
BackgroundSex estimation based on deciduous and permanent dentition in a contemporary spanish population
Viciano, J., López‐Lázaro, S., & Aguilera, I. A. (2013) · American Journal of Physical Anthropology · doi:10.1002/ajpa.22324 · Cited by 114
Kirol, C. P., Beck, J. L., Huzurbazar, S. V., Holloran, M. J., & Miller, S. N. (2014) · Ecological Applications · doi:10.1890/13-1152.1 · Cited by 96
Long-Term Outcome and its Predictors in Giant Cell Myocarditis
Ekström, K., Lehtonen, J., Kandolin, R., Räisänen-Sokolowski, A., Salmenkivi, K., & Kupari, M. (2016) · European Journal of Heart Failure · doi:10.1002/ejhf.606 · Cited by 91
“Immunosuppressive therapya 37/38 (97) Prednisone 37/37 (100) Azathioprine 31/37 (84) Cyclosporine 28/37 (76) Otherb 7/37 (19) Triple combination therapyc 26/37 (70) Beta-adrenergic blockers 39 (85) ACE inhibitors 33 (72) Amiodarone 28 (61) ICD implanted (total) 26 (57) Primary prevention 21 (46) Secondary prevention 5…”
BackgroundStewart, J. A. E., Perrine, J. D., Nichols, L. B., Thorne, J. H., Millar, C. I., Goehring, K. E., Massing, C. P., & Wright, D. H. (2015) · Journal of Biogeography · doi:10.1111/jbi.12466 · Cited by 96
“To avoid over-fitting we restricted all models to a maximum of five events per variable (Vittinghoff & McCulloch, 2007).”
MethodSimple and multiple linear regression: sample size considerations
Hanley, J. A. (2016) · Journal of Clinical Epidemiology · doi:10.1016/j.jclinepi.2016.05.014 · Cited by 107
Hirst, J. E., Villar, J., Victora, C. G., Papageorghiou, A. T., Finkton, D. W., Barros, F. C., Gravett, M., Giuliani, F., Purwar, M., Frederick, I., et al. (2016) · BJOG An International Journal of Obstetrics & Gynaecology · doi:10.1111/1471-0528.14463 · Cited by 261
Lin, X., Zhang, D., & Li, Y. (2016) · Computers in Human Behavior · doi:10.1016/j.chb.2016.01.017 · Cited by 118
Wenzel, H. H. B., Smolders, R. G. V., Beltman, J. J., Lambrechts, S., Trum, H. W., Yigit, R., Zusterzeel, P. L. M., Zweemer, R. P., Mom, C. H., Bekkers, R. L. M., et al. (2020) · European Journal of Cancer · doi:10.1016/j.ejca.2020.04.006 · Cited by 96
“In addition, at least 5 events per predictor parameter in multivariable analysis were required to prevent model overfitting [13] or, alternatively, a propensity score [14].”
BackgroundLoymans, R. J. B., Honkoop, P., Termeer, E. H., Snoeck-Stroband, J. B., Assendelft, W. J., Schermer, T. R., Chung, K. F., Sousa, A. R., Sterk, P. J., Reddel, H. K., et al. (2016) · Thorax · doi:10.1136/thoraxjnl-2015-208138 · Cited by 97
Phillippo, D., Dias, S., Ades, A. E., & Welton, N. J. (2020) · Statistics in Medicine · doi:10.1002/sim.8759 · Cited by 97
Djärv, T., Metcalfe, C. D., AVERY, K. N. L., Lagergren, P., & Blazeby, J. M. (2010) · Journal of Clinical Oncology · doi:10.1200/jco.2009.23.5143 · Cited by 88
Serum sclerostin is an independent predictor of mortality in hemodialysis patients
Gonçalves, F. L. C., Elias, R. M., dos Reis, L. M., Graciolli, F. G., Zampieri, F. G., de Oliveira, R. C., Jorgetti, V., & Moysés, R. M. A. (2014) · BMC Nephrology · doi:10.1186/1471-2369-15-190 · Cited by 88
“Finally, we used an event per variable ratio that may be considered low for regression [26], which might be associated with overfitting.”
MethodO’Leary, H., Smart, K. M., Moloney, N. A., Blake, C. M., & Doody, C. (2018) · Pain · doi:10.1097/j.pain.0000000000001288 · Cited by 95
Potter, B. J., Leong‐Sit, P., Fernandes, S. M., Feifer, A. H., Mayer, J. E., Triedman, J. K., Walsh, E. P., Landzberg, M. J., & Khairy, P. (2013) · International Journal of Cardiology · doi:10.1016/j.ijcard.2013.06.058 · Cited by 87
“Moreover, simulation studies suggest that problems are seldom encountered with 5–9 events per predictor variable [34].”
BackgroundVedder, M. M., de Bekker-Grob, E. W., Lilja, H. G., Vickers, A. J., van Leenders, G. J. L. H., Steyerberg, E. W., & Roobol, M. J. (2014) · European Urology · doi:10.1016/j.eururo.2014.08.011 · Cited by 94
“Ideally the EPV would be higher, but EPV values from 5 have been 247 shown to be valid in the context of statistical adjustment for baseline risk factors (31).”
The natural history of compensated HCV-related cirrhosis: A prospective long-term study
Gómez, E. V., Rodriguez, Y. S., Bertot, L. C., Gonzalez, A. T., Pérez, Y. M., Soler, E. A., Garcia, A. Y., & Blanco, L. P. (2012) · Journal of Hepatology · doi:10.1016/j.jhep.2012.10.023 · Cited by 91
Transtibial ACL Femoral Tunnel Preparation Increases Odds of Repeat Ipsilateral Knee Surgery
Duffee, A. R., Magnussen, R. A., Pedroza, A. D., Flanigan, D. C., MOON Group, & Kaeding, C. C. (2013) · Journal of Bone and Joint Surgery · doi:10.2106/jbjs.m.00187 · Cited by 90
Lim, G., Farrell, L. M., Facco, F. L., Gold, M. S., & Wasan, A. D. (2018) · Anesthesia & Analgesia · doi:10.1213/ane.0000000000002720 · Cited by 84
Analyzing the Influence of Micro-Level Factors on CCTV Camera Effect
Piza, E. L., Caplan, J. M., & Kennedy, L. W. (2013) · Journal of Quantitative Criminology · doi:10.1007/s10940-013-9202-5 · Cited by 98
Influence of pneumococcal serotype group on outcome in adults with bacteraemic pneumonia
Luján, M., Gallego, M., Belmonte, Y., Fontanals, D., Vallés, J., Lisboa, T. C., & Rello, J. (2010) · European Respiratory Journal · doi:10.1183/09031936.00176309 · Cited by 83
“of at least 5-7 events (deaths) per variable [19].”
BackgroundHigher Frequency of Reoperation With a New Bicruciate-retaining Total Knee Arthroplasty
Christensen, J. C., Brothers, J. G., Stoddard, G. J., Anderson, M. B., Pelt, C. E., Gililland, J. M., & Peters, C. L. (2016) · Clinical Orthopaedics and Related Research · doi:10.1007/s11999-016-4812-5 · Cited by 87
“It has been shown that five events for each predictor in Cox regression is sufficient to avoid overfitting when the aim is to adjust for confounding [28], so given two predictors, group and BMI, we needed 10 events.”
BackgroundOlsen, F. J., Møgelvang, R., Jensen, G. B., Jensen, J. S., & Biering‐Sørensen, T. (2018) · JACC. Cardiovascular imaging · doi:10.1016/j.jcmg.2017.12.016 · Cited by 90
“However, this approach can be acceptable when the focus is on controlling for confounders rather than building prediction models (25).”
BackgroundA greedy feature selection algorithm for Big Data of high dimensionality
Tsamardinos, I., Borboudakis, G., Katsogridakis, P., Pratikakis, P., & Christophides, V. (2018) · Machine Learning · doi:10.1007/s10994-018-5748-7 · Cited by 104
Nieto-Castañón, A. (2022) · PLoS Computational Biology · doi:10.1371/journal.pcbi.1010634 · Cited by 89
Pets, Depression and Long-term Survival in Community Living Patients Following Myocardial Infarction
Friedmann, E., Thomas, S. A., & Son, H. (2011) · Anthrozoös · doi:10.2752/175303711x13045914865268 · Cited by 87
“Recent simulations indicate that 5 to 9 deaths per predictor yield reliable results in logistic and Cox regression (Vittinghoff and McCulloch 2007).”
BackgroundBrousseau, A.-A., Dent, E., Hubbard, R. E., Melady, D., Émond, M., Mercier, É., Costa, A. P., Gray, L., Hirdes, J. P., Dey, A. B., et al. (2017) · Age and Ageing · doi:10.1093/ageing/afx168 · Cited by 88
“The number of events included in each cohort provided more than adequate power to estimate associations between the FI-ED and adverse outcomes [23].”
BackgroundRamos, L. A., van der Steen, W. E., Barros, R. S., Majoie, C. B. L. M., van den Berg, R., Verbaan, D., Vandertop, W. P., Zijlstra, I. A. J., Zwinderman, A. H., Strijkers, G. J., et al. (2018) · Journal of NeuroInterventional Surgery · doi:10.1136/neurintsurg-2018-014258 · Cited by 84
Arguelles, O. C., Abdel-Qadir, H. M., Michalowska, M., Billia, F., Suntheralingam, S., Amir, E., & Thavendiranathan, P. (2018) · Canadian Journal of Cardiology · doi:10.1016/j.cjca.2018.11.028 · Cited by 95
Ziermans, T. B., de Wit, S., Schothorst, P. F., Sprong, M., van Engeland, H., Kahn, R. S., & Durston, S. (2014) · PLoS ONE · doi:10.1371/journal.pone.0093994 · Cited by 89
“Ideally, regression analyses include 10 events per predictor variable or more, although smaller numbers can still produce robust results, albeit with a greater risk of introducing bias [51].”
BackgroundUsing citizen science data to estimate climatic niches and species distributions
Tiago, P., Pereira, H. M., & Capinha, C. (2017) · Basic and Applied Ecology · doi:10.1016/j.baae.2017.04.001 · Cited by 97
Flynn, R. W. V. (2012) · Journal of Clinical Nursing · doi:10.1111/j.1365-2702.2011.04023.x · Cited by 91
Watkins, D. J., Sánchez, B. N., Téllez‐Rojo, M. M., Lee, J. M., Mercado‐García, A., Blank-Goldenberg, C., Peterson, K. E., & Meeker, J. D. (2017) · Environmental Health · doi:10.1186/s12940-017-0278-5 · Cited by 90
“As few participants were categorized as having a PH Tanner stage >1 (Additional file 1: Table S2), the number of covariates we could reliably enter into models was limited [25, 26].”
BackgroundDog ownership satisfaction determinants in the owner-dog relationship and the dog's behaviour
van Herwijnen, I. R., van der Borg, J. A. M., Naguib, M., & Beerda, B. (2018) · PLoS ONE · doi:10.1371/journal.pone.0204592 · Cited by 104
Representing their own? Ethnic minority women in the Dutch Parliament
Mügge, L. M., van der Pas, D., & van de Wardt, M. (2019) · West European Politics · doi:10.1080/01402382.2019.1573036 · Cited by 116
Nam, J. L., Hunt, L. H., Hensor, E. M. A., & Emery, P. (2015) · Annals of the Rheumatic Diseases · doi:10.1136/annrheumdis-2015-207871 · Cited by 79
Allou, N., Pinto, H. L., Persichini, R., Bouchet, B., Braunberger, É., Lugagne, N., Belmonte, O., Martinet, O., Delmas, B., Dangers, L., et al. (2018) · ASAIO Journal · doi:10.1097/mat.0000000000000771 · Cited by 79
Mixed‐effects models for slope‐based endpoints in clinical trials of chronic kidney disease
Vonesh, E., Tighiouart, H., Ying, J., Heerspink, H. L., Lewis, J., Staplin, N., Inker, L., & Greene, T. (2019) · Statistics in Medicine · doi:10.1002/sim.8282 · Cited by 78
“Statistics in Medicine, 38(22), 4218-4239.”
Du, M., Haag, D. G., Lynch, J. W., & Mittinty, M. (2020) · Cancers · doi:10.3390/cancers12102802 · Cited by 95
Skandsen, T., Kvistad, K. A., Solheim, O. S., Lydersen, S., Strand, I. H., & Vik, A. (2011) · Journal of Neurotrauma · doi:10.1089/neu.2010.1590 · Cited by 83
“In the binary logistic regression analysis, the number of covariates was limited to a maximum of one-fifth of the number of events in the smallest group (Vittinghoff and McCulloch, 2007).”
Atrial fibrillation type and renal dysfunction as important predictors of left atrial thrombus
Kapłon‐Cieślicka, A., Budnik, M., Gawalko, M. M., Peller, M., Gorczyca−Michta, I., Michalska, A., Babiarz, A., Bodys-Pełka, A., Ulińśki, R., Żochowski, M., et al. (2019) · Heart · doi:10.1136/heartjnl-2018-314492 · Cited by 84
“Multivariate logistic regression model included all variables found to be predictors of LAA thrombus in univariate analyses, maintaining an adequate event per predictor variable value.(8) A second multivariate model was derived, including the CHA2DS2-VASc score, to enable adequate point attribution.”
MethodModin, D., Møgelvang, R., Andersen, D. M., & Biering‐Sørensen, T. (2019) · Journal of the American Heart Association · doi:10.1161/jaha.119.012197 · Cited by 86
Eisenmenger, L. B., Aldred, B., Kim, S. E., Stoddard, G. J., de Havenon, A., Treiman, G. S., Parker, D. L., & McNally, J. S. (2016) · American Journal of Neuroradiology · doi:10.3174/ajnr.a4765 · Cited by 75
“outcome events for every predictor variable are sufficient to avoid overfitting.(30) With 44 carotid IPH events and 144 non-IPH events, 44/5 8.”
BackgroundJu, K. L., Zurakowski, D., & Kocher, M. S. (2011) · Journal of Bone and Joint Surgery · doi:10.2106/jbjs.j.01154 · Cited by 76
Duwe, G., & Clark, V. A. (2015) · Criminology & Public Policy · doi:10.1111/1745-9133.12123 · Cited by 93
Hay, A. D., Redmond, N. M., Turnbull, S., Christensen, H., Thornton, H., Little, P. S., Thompson, M., Delaney, B. C., Lovering, A. M., Muir, P., et al. (2016) · The Lancet Respiratory Medicine · doi:10.1016/s2213-2600(16)30223-5 · Cited by 82
The national and subnational prevalence and burden of age–related macular degeneration in China
Song, P., Du, Y., Chan, K. Y., Τheodoratou, E., & Rudan, I. (2017) · Journal of Global Health · doi:10.7189/jogh.07.020703 · Cited by 85
Téllez, L., de Santiago, E. R., Mínguez, B., Payancé, A., Clemente, A. P. G., Baiges, A., Morales-Arráez, D. E., Mura, V. L., Llop, E., Garrido, E., et al. (2019) · Journal of Hepatology · doi:10.1016/j.jhep.2019.10.027 · Cited by 76
“For multiva riate analyses, the rule of a minimum of ten events per variable was used [15].”
MethodWinquist, A., Hodge, J. M., Diver, W. R., Rodriguez, J. L., Troeschel, A. N., Daniel, J., & Teras, L. R. (2023) · Environmental Health Perspectives · doi:10.1289/ehp13174 · Cited by 75
“113 Models for bladder, breast, prostate, and the larger hematologic malignancy subgroups had an ample number of cases relative to the number of parameters, but models for kidney cancer, pancreatic cancer and smaller hematologic sub-types had a fewer cases per parameter (Table S11).”
BackgroundJohnson-Ulrich, L., Johnson‐Ulrich, Z., & Holekamp, K. E. (2018) · Animal Cognition · doi:10.1007/s10071-018-1174-2 · Cited by 83
“Therefore, null results must be interpreted cautiously, but significant results can be interpreted normally, albeit with a degree of caution regarding generality (Vittinghoff and McCulloch 2007).”
BackgroundFuest, K. E., Ulm, B., Daum, N., Lindholz, M., Lorenz, M., Blobner, K., Langer, N., Hodgson, C. L., Herridge, M. S., Blobner, M., et al. (2023) · Critical Care · doi:10.1186/s13054-022-04291-8 · Cited by 91
“Power analysis According to Vittinghoff and McCulloch, a case number of 5–9 patients per variable with the smaller outcome category is required to ensure sufficient power [35].”
MethodTrotti, L. M., Rye, D. B., Staercke, C. D., Hooper, W. C., Quyyumi, A. A., & Bliwise, D. L. (2012) · Brain Behavior and Immunity · doi:10.1016/j.bbi.2012.06.003 · Cited by 79
“Our overall sample size was small relative to the number of potential confounders evaluated in the multivariate model, which may have affected the precision of the odds ratio estimates (Vittinghoff and McCulloch, 2007).”
BackgroundHoffmayer, K. S., Machado, O. N., Marcus, G. M., Yang, Y., Johnson, C. J., Ermakov, S., Vittinghoff, E., Pandurangi, U. M., Calkins, H. G., Cannom, D. S., et al. (2011) · Journal of the American College of Cardiology · doi:10.1016/j.jacc.2011.05.017 · Cited by 77
“4 6 rule of 1 covariate per 10 outcomes, although a recent paper suggests that this criterion may be too stringent (19); however, we acknowledge that our ratio is quite small.”
BackgroundSelf-citationThe Quality of Modern Cross-Sectional Ecologic Studies: A Bibliometric Review
Dufault, B., & Klar, N. (2011) · American Journal of Epidemiology · doi:10.1093/aje/kwr241 · Cited by 73
“The most common recommendation is to have a minimum of 10 observations per covariate, though this may be too conservative for some models (11).”
Backgroundvan den Berg, E., Pasch, A., Westendorp, W. H., Navis, G. J., Brink, E. J., Gans, R. O. B., van Goor, H., & Bakker, S. J. L. (2014) · Journal of the American Society of Nephrology · doi:10.1681/asn.2013050497 · Cited by 76
Oral health care behavior and frailty-related factors in a care-dependent older population
Niesten, D., Witter, D. J., Bronkhorst, E. M., & Creugers, N. H. J. (2017) · Journal of Dentistry · doi:10.1016/j.jdent.2017.04.002 · Cited by 86
Stronger social bonds do not always predict greater longevity in a gregarious primate
González, N. T., & Cords, M. (2018) · Ecology and Evolution · doi:10.1002/ece3.3781 · Cited by 83
“Survival models have sufficient power when each variable corresponds with 5—10 events (Vittinghoff & McCulloch, 2007).”
BackgroundGut Microbiota-Derived Metabolite Trimethylamine N-Oxide as a Biomarker in Early Parkinson’s Disease
Chung, S. J., Rim, J. H., Ji, D., Lee, S., Yoo, H. S., Jung, J. H., Baik, K., Choi, Y., Ye, B. S., Sohn, Y. H., et al. (2021) · Nutrition · doi:10.1016/j.nut.2020.111090 · Cited by 74
Treatment challenges associated with bone echinococcosis
Steinmetz, S., Racloz, G., STERN, R. E., Dominguez, D. E., Al‐Mayahi, M., Schibler, M., Lew, D., Hoffmeyer, P. J., & Uçkay, İ. (2013) · Journal of Antimicrobial Chemotherapy · doi:10.1093/jac/dkt429 · Cited by 74
Aronson, S., Dyke, C., Levy, J. H., Cheung, A. T., Lumb, P. D., Avery, E. G., Hu, M., & Newman, M. F. (2011) · Anesthesia & Analgesia · doi:10.1213/ane.0b013e31820f9231 · Cited by 74
Glangeaud-Freudenthal, N. M.-C., Sutter, A.-L., Thieulin, A., Dagens-Lafont, V., Zimmermann, M., Debourg, A., Massari, B., Cazas, O., Cammas, R., Rainelli, C., et al. (2010) · European Psychiatry · doi:10.1016/j.eurpsy.2010.03.006 · Cited by 76
“Specifically, the literature recommends a minimum of five to 10 events per variable [29,44].”
BackgroundBest, K. E., Tennant, P. W. G., & Rankin, J. (2017) · Journal of the American Heart Association · doi:10.1161/jaha.116.005213 · Cited by 74
Lunn, T. H., Gaarn-Larsen, L., & Kehlet, H. (2013) · Pain · doi:10.1016/j.pain.2013.06.008 · Cited by 73
“However, we consider the sample size to be sufficient as to both the multivariate linear regression analysis with each of the 3 outcomes (11 to 12 observations were present for each of the 8 explanatory variables, and about 10 are generally recommended) [5], and the multivariate logistic regression analysis (36 patien…”
MethodMurray-Thomas, T., Jones, M. E., Patel, D., Brunner, E., Shatapathy, C. C., Motsko, S., & Van Staa, T. P. (2013) · Cardiovascular Psychiatry and Neurology · doi:10.1155/2013/247486 · Cited by 72
“To ensure that the resulting models were not overparameterised, the number of variables to be included in the model was selected using the rule of thumb of five outcome events per parameter [21, 22].”
MethodA 6-year longitudinal study of predictors for suicide attempts in major depressive disorder
Eikelenboom, M., Beekman, A. T. F., Penninx, B. W. J. H., & Smit, J. H. (2019) · Psychological Medicine · doi:10.1017/s0033291718001423 · Cited by 71
“Moreover, a simulation study showed that systematically discounting results from any model with 5–9 EVP does not appear to be justified (Vittinghof and McCulloch, 2006).”
BackgroundCyber-Routines, Political Attitudes, and Exposure to Violence-Advocating Online Extremism
Hawdon, J., Bernatzky, C., & Costello, M. (2019) · Social Forces · doi:10.1093/sf/soy115 · Cited by 72
“…with 230 individuals being exposed to violence-advocating materials in our sample, the analysis is well within the general “ ten outcome events per variable ” (EPV) rule, especially given the need to control for potential confounding variables (see Harrell 2015; Vittinghoff and McCulloch 2007).”
BackgroundVardy, T., & Atkinson, Q. D. (2019) · Psychological Science · doi:10.1177/0956797619826972 · Cited by 71
“…the widely cited rule of thumb that one should obtain at least 10 data points per predictor variable (Concato, Peduzzi, Holford, & Feinstein, 1995; Vittinghoff & McCulloch, 2007) suggests that our sample was sufficiently powered for the main analyses we report (with between six and eight…”
BackgroundKoskela, M., Ylinen, E., Ukonmaanaho, E., Autio‐Harmainen, H., Heikkilä, P., Lohi, J., Jauhola, O., Ronkainen, J., Jahnukainen, T., & Nuutinen, M. (2017) · Pediatric Nephrology · doi:10.1007/s00467-017-3608-5 · Cited by 71
A rapid, non-invasive tool for periodontitis screening in a medical care setting
Verhulst, M. J. L., Teeuw, W. J., Bizzarro, S., Muris, J., Su, N., Nicu, E. A., Nazmi, K., Bikker, F. J., & Loos, B. G. (2019) · BMC Oral Health · doi:10.1186/s12903-019-0784-7 · Cited by 79
Schmidt, J., Chiffelle, J., Perez, M. A. S., Magnin, M., Bobisse, S., Arnaud, M., Genolet, R., Cesbron, J., Barras, D., Rodrigo, B. N., et al. (2023) · Nature Communications · doi:10.1038/s41467-023-38946-z · Cited by 84
“Combinations of 5 – 8 amino acids were explored to alleviate over fi tting thanks to 5 – 9 TCRs per explanatory variable 75 .”
Yancey, G. (2009) · Sociological Quarterly · doi:10.1111/j.1533-8525.2008.01135.x · Cited by 80
“In Asian models, the EPV is only 8, but Vittinghoff and McCulloch (2007) suggest EPV scores in the range 5 to 9 do not usually generate problematic analysis.”
BackgroundSarcopenia in Advanced Serous Ovarian Cancer
Bronger, H., Hederich, P., Hapfelmeier, A., Metz, S., Noël, P. B., Kiechle, M. B., & Schmalfeldt, B. (2016) · International Journal of Gynecological Cancer · doi:10.1097/igc.0000000000000867 · Cited by 75
“With event counts of 65 and 40 for the progression-free and overall survival models, respectively, we limited the size of our multivariable to 5 estimated HRs to avoid overspecification issues and to ensure consistent parameter estimation.(25,26) In this model, sarcopenia at baseline remained significantly predictive…”
MethodDharma, C., Lefebvre, D. L., Tran, M. M., Lou, W. Y. W., Subbarao, P., Becker, A. B., Mandhane, P. J., Turvey, S. E., Sears, M. R., & the CHILD Study investigators (2017) · Clinical & Experimental Allergy · doi:10.1111/cea.13063 · Cited by 81
“2, demographic factors that were associated with both outcomes and predictor variables were included as adjustment variables, while ensuring that enough events were observed given the number of covariates.(31) To calculate the robust variance while accounting for the regression weights, the software STATA 13 was used…”
MethodBrown, M. M., Cummings, C., Lyons, J., Carrión, A., & Watson, D. P. (2018) · Journal of Social Distress and the Homeless · doi:10.1080/10530789.2018.1482991 · Cited by 83
“Twenty-seven individuals re-entered services, which was adequate for the models (Vittinghoff & McCulloch, 2007).”
MethodMetachronous colorectal cancers result from missed lesions and non-compliance with surveillance
le Clercq, C. M. C., Winkens, B., Bakker, C. M., Keulen, E. T. P., Beets, G. L., Masclee, A., & Sanduleanu, S. (2015) · Gastrointestinal Endoscopy · doi:10.1016/j.gie.2014.12.052 · Cited by 72
Viasus, D., Paño‐Pardo, J. R., Cordero, E., Campins, A. A., López‐Medrano, F., Villoslada, A., Fariñas, M. C., Moreno, A., Rodríguez‐Baño, J., Oteo, J. A., et al. (2011) · Journal of Infection · doi:10.1016/j.jinf.2011.01.014 · Cited by 69
van der Linden, M., Falkenhorst, G., Perniciaro, S., Fitzner, C., & Imöhl, M. (2016) · PLoS ONE · doi:10.1371/journal.pone.0161257 · Cited by 75
“Another limitation to our study, and indeed in many other published studies using logistic regression to estimate odds ratios for VE calculations, is that low event numbers (at least 5–10 events per predictor variable are recommended [26, 27]), which commonly occur in the rapidly-disappearing vaccine-type serotypes, s…”
ResultBinary Response Analysis Using Logistic Regression in Dentistry
Srimaneekarn, N., Hayter, A. J., Liu, W., & Tantipoj, C. (2022) · International Journal of Dentistry · doi:10.1155/2022/5358602 · Cited by 86
“However, if the sample size is limited, Vittinghoff et al. (2007) stated that 5–9 events per variable with bootstrap resampling validation was acceptable [10].”
Signs and symptoms of temporomandibular disorders and the incidence of tinnitus
Bernhardt, O., Mündt, T., Welk, A., KÖPPL, N., Kocher, T., Meyer, G. F., & Schwahn, C. (2011) · Journal of Oral Rehabilitation · doi:10.1111/j.1365-2842.2011.02224.x · Cited by 75
“The difference between ‘frequent’ and ‘always’ tinnitus, however, was already present in the model adjusted only for age and sex which fulfilled the rule of five events per variable (31).”
BackgroundThe Effect of Osteoporosis on Outcomes of Operatively Treated Distal Radius Fractures
FitzPatrick, S. K., Casemyr, N., Zurakowski, D., Day, C. S., & Rozental, T. D. (2012) · The Journal Of Hand Surgery · doi:10.1016/j.jhsa.2012.06.025 · Cited by 72
Choice of birth control methods among European women and the role of partners and providers
de Irala, J., Osorio, A., Carlos, S., & del Burgo, C. L. (2011) · Contraception · doi:10.1016/j.contraception.2011.04.004 · Cited by 73
“According to the criterion that 10 subjects would be needed for each parameter included in a statistical model to adjust for confounding [13], our global sample size of 1137 was sufficient for these purposes.”
BackgroundMethodA blood-based biomarker panel to risk-stratify mild traumatic brain injury
Sharma, R., Rosenberg, A., Bennett, E. R., Laskowitz, D. T., & Acheson, S. K. (2017) · PLoS ONE · doi:10.1371/journal.pone.0173798 · Cited by 71
Carroll, B., Beyer, S. E., Mehegan, T., Dicks, A. B., Pribish, A. M., Locke, A. H., Godishala, A., Soriano, K. M., Kanduri, J. S., Sack, K. D., et al. (2020) · The American Journal of Medicine · doi:10.1016/j.amjmed.2020.03.058 · Cited by 71
Obesity Is Associated with Risk of Progression for Low-risk Prostate Cancers Managed Expectantly
Bhindi, B., Kulkarni, G. S., Finelli, A., Alibhai, S. M. H., Hamilton, R. J., Toi, A., van der Kwast, T., Evans, A. J., Hersey, K. M., Jewett, M. A. S., et al. (2014) · European Urology · doi:10.1016/j.eururo.2014.06.005 · Cited by 67
“Although the Cox models did not have 10 events per degree of freedom, there is evidence to suggest that this guideline is overly conservative [13].”
MethodPollet, T. V., Roberts, S. G. B., & Dunbar, R. (2013) · PLoS ONE · doi:10.1371/journal.pone.0053929 · Cited by 77
“However, a low frequency likely would increase Type 1 errors making it harder to find the results detailed above, and simulations suggest that Cox regression performs relatively well even when the number of events are low [81].”
BackgroundIkdahl, E., Rollefstad, S., Wibetoe, G., Olsen, I. C., Berg, I.-J., Hisdal, J., Uhlig, T., Haugeberg, G., Kvien, T. K., Provan, S. A., et al. (2016) · The Journal of Rheumatology · doi:10.3899/jrheum.160053 · Cited by 71
Ngamruengphong, S., Inoue, H., Chiu, P. W. Y., Yip, H. C., Bapaye, A., Ujiki, M. B., Patel, L. Y., Desai, P. N., Hayee, B., Haji, A., et al. (2016) · Gastrointestinal Endoscopy · doi:10.1016/j.gie.2016.09.017 · Cited by 67
RetractedRemiro‐Azócar, A., Heath, A., & Baio, G. (2021) · Research Synthesis Methods · doi:10.1002/jrsm.1511 · Cited by 81
The Decision to Exploit Opportunities for Innovation: A Study of High–Tech Small–Business Owners
de Jong, J. P. J. (2012) · Entrepreneurship Theory and Practice · doi:10.1111/j.1540-6520.2011.00459.x · Cited by 91
Kovacs, A., Vadeboncoeur, T. F., Stolz, U., Spaite, D. W., Irisawa, T., Silver, A., & Bobrow, B. J. (2015) · Resuscitation · doi:10.1016/j.resuscitation.2015.04.026 · Cited by 67
Davila, A. A., Seth, A. K., Wang, E., Hanwright, P. J., Bilimoria, K. Y., Fine, N. A., & Kim, J. Y. (2013) · Archives of Plastic Surgery · doi:10.5999/aps.2013.40.1.19 · Cited by 67
“Additionally, variables that did not have greater than 5 events were excluded from further analysis [21].”
BackgroundFleury, T. R., Dunkel, N., Assal, M., Rohner, P., Tahintzi, P., Bernard, L., Hoffmeyer, P. J., Lew, D., & Uçkay, İ. (2011) · International Orthopaedics · doi:10.1007/s00264-011-1221-y · Cited by 67
“We included six to ten predictor variables per outcome event [ 20 ].”
Metadata
- Authors
- Vittinghoff, E., & McCulloch, C. E.
- Affiliations
- University of California, San Francisco (US)
- Venue
- American Journal of Epidemiology
- Publisher
- Oxford University Press (OUP)
- Published
- 2007-01-12
- Volume
- 165
- Issue
- 6
- Pages
- 710-718
- ISSN
- 0002-9262 (ISSN-L)
- 1476-6256
- OpenAlex
- W2130373985
- Cited by
- 4,359
- References
- 9
- Citations by year
- 2026: 642 · 2025: 394 · 2024: 313 · 2023: 315 · 2022: 306 · 2021: 351 · 2020: 332 · 2019: 281 · 2018: 282 · 2017: 205 · 2016: 194 · 2015: 171 · 2014: 146 · 2013: 132 · 2012: 94
- FWCI
- 20.12
- Type
- journal article
- Language
- en
- Open access
- closed
- Locations
- American Journal of Epidemiology (publishedVersion)
- PubMed (publishedVersion)
- Topics
- Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods in Epidemiology
- Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods and Inference
- Physical Sciences › Mathematics › Statistics and Probability › Advanced Causal Inference Techniques
- Keywords
- sample size, logistic regression, Cox proportional hazards model, confidence interval coverage, type I error, relative bias, simulation study, sensitivity analysis, confounding adjustment, small sample size
- MeSH terms
- Data Interpretation, Statistical, Factor Analysis, Statistical, Humans, Predictive Value of Tests, Sensitivity and Specificity, Bias, Confounding Factors, Epidemiologic, Prevalence, Multivariate Analysis, Confidence Intervals, Logistic Models, Proportional Hazards Models, Case-Control Studies, Sample Size
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.