Inflation of the type I error rate when a continuous confounding variable is categorized in logistic regression analyses

Austin, Peter C.; Brunner, Lawrence J.

doi:10.1002/sim.1687

Cited by 197 publications

(124 citation statements)

References 24 publications

Supporting

Mentioning

116

Contrasting

Unclassified

Order By: Relevance

“…Stratification by categories (e.g., 5 categories) of the confounder may adequately control for the confounding by continuous confounders as well, 17 although in extreme settings this may not hold. 21 In our clinical examples, control for continuous confounders was indeed similar when stratifying using 5 strata or when using fractional polynomials or restricted cubic splines. It should be noted that the results from these methods differed compared with adjustment by stratifying on the dichotomized continuous confounder.…”

Section: Discussionmentioning

confidence: 67%

Adjustment for continuous confounders: an example of how to prevent residual confounding

Groenwold

Klungel

Altman

et al. 2013

CMAJ

View full text Add to dashboard Cite

Analysis CMAJ• Adjustment for confounding is crucial in observational studies on etiology or on preventive or therapeutic treatment effects.• A review of current practice in the reporting of adjustment for continuous confounders in observational epidemiologic studies showed that the functional relation between continuous confounders and outcome is hardly ever reported.• Incorrect modelling of a continuous confounder can result in important residual confounding, as shown by clinical examples.• In these clinical examples, adjustment for continuous confounding by means of stratification of the confounder in 5 strata, and use of fractional polynomials or restricted cubic splines yielded similar results.

show abstract

Section: Discussionmentioning

confidence: 67%

Adjustment for continuous confounders: an example of how to prevent residual confounding

Groenwold

Klungel

Altman

et al. 2013

CMAJ

View full text Add to dashboard Cite

show abstract

“…There was no evidence to suggest that data were normally distributed, hence in the descriptive statistics for continuous variables, we report median and inter-quartile range. To avoid inflating the type-I error rate, loss of power, residual confounding, and bias, continuous predictor variables were not categorised (Del Priore et al, 1997;Austin and Brunner, 2004;Royston et al, 2006). To test any differences we used a non-parametric Wilcoxon rank sum (Mann -Whitney) test.…”

Section: Discussionmentioning

confidence: 99%

Predicting the risk of endometrial cancer in postmenopausal women presenting with vaginal bleeding: the Norwich DEFAB risk assessment tool

et al. 2010

View full text Add to dashboard Cite

BACKGROUND: This study aimed to show the longitudinal use of routinely collected clinical data from history and ultrasound evaluation of the endometrium in developing an algorithm to predict the risk of endometrial carcinoma for postmenopausal women presenting with vaginal bleeding. METHODS: This prospective study collected data from 3047 women presenting with postmenopausal bleeding. Data regarding the presence of risk factors for endometrial cancer was collected and univariate and multivariate analyses were performed. RESULTS: Age distribution ranged from 35 to 97 years with a median of 59 years. A total of 149 women (5% of total) were diagnosed with endometrial carcinoma. Women in the endometrial cancer group were significantly more likely to be older, have higher BMI, recurrent episodes of bleeding, diabetes, hypertension, or a previous history of breast cancer. An investigator best model selection approach was used to select the best predictors of cancer, and using logistic regression analysis we created a model, 'Norwich DEFAB', which is a clinical prediction rule for endometrial cancer. The calculated Norwich DEFAB score can vary from a value of 0 to 9. A Norwich DEFAB value equal to or greater than 3 has a positive predictive value (PPV) of 7.78% and negative predictive value (NPV) of 98.2%, whereas a score equal to or greater than 5 has a PPV of 11.9% and NPV of 97.8%. CONCLUSION: The combination of clinical information with our investigation tool for women with postmenopausal vaginal bleeding allows the clinician to calculate a predicted risk of endometrial malignancy and prioritise subsequent clinical investigations.

show abstract

“…Dichotomization is very common in clinical studies, since it satisfies the general need in clinical practice to label individuals as having or not having a characteristic and to decide the treatment strategy. However, converting all patients into two categories has some intrinsic problems, such as loss of information and statistical power, increased probability of false positive results, and impossibility of detecting the shape of the relationship [15][16][17][18]. In addition, in the absence of a priori information, ''optimal'' cutpoints are usually obtained by trying many thresholds and choosing the one which, to some extent, gives the most satisfactory result, for example, choosing the value which minimizes the P-value of the Log-rank test [13].…”

Section: Discussionmentioning

confidence: 99%

“…For example, in studies on MBC, a cutpoint of 5 CTCs has been generally adopted [3,4,13,14]. Dichotomization has some perceived advantages, such as simplicity, easy interpretation of the results, and functionality in clinical practice, but also some intrinsic problems, such as loss of information and statistical power, increased probability of false positive results, and impossibility of detecting nonlinear relationships between the predictor and the outcome [15][16][17][18].…”

Section: Introductionmentioning

confidence: 99%

Modeling the relationship between circulating tumour cells number and prognosis of metastatic breast cancer

Botteri

Sandri

Bagnardi

et al. 2009

Breast Cancer Res Treat

View full text Add to dashboard Cite

Modeling the relationship between circulating tumour cells number and prognosis of metastatic breast cancer. Breast Cancer Research and Treatment, Springer Verlag, 2009, 122 (1), pp.211-217. <10.1007/s10549-009-0668-7>. EPIDEMIOLOGYModeling the relationship between circulating tumour cells number and prognosis of metastatic breast cancer Abstract Circulating tumor cell (CTC) count has been shown to be an independent predictor of progression in metastatic breast, prostate, and colorectal cancer. A cutpoint is generally used to identify favorable and unfavorable response groups. In this study, we propose an approach in which the number of CTCs is analyzed as a continuous predictor, to detect the shape of the relationship between CTCs and prognosis of metastatic breast cancer. We evaluated the association of baseline CTC with progression-free survival (PFS) and overall survival (OS) in a series of 80 patients treated for advanced breast cancer at the European Institute of Oncology, Milan. The association between CTCs and prognosis was analyzed with standard categorical survival analysis and spline regression models. At baseline, median age was 55 years; 33 patients were newly diagnosed with metastatic breast cancer (41%), while 28 (35%) and 19 (24%) were pretreated with one and two previous chemotherapy lines, respectively. After a median follow-up of 28 months, 76 disease progressions and 44 deaths were observed. Kaplan-Meier curves showed a clear association between CTCs and PFS (P-value 0.03) and OS (Pvalue \ 0.01). Patients with no CTC at baseline had a significantly better prognosis. When analyzing the CTCs as a continuous variable, we found an increase in risk with increasing number of CTCs, for both PFS and OS. The increase rate lessened after approximately 5 CTCs. CTCs represent a robust prognostic factor in the metastatic breast cancer setting. A nonlinear increase in risk of both progression and death with increasing number of CTCs was observed, with a lessening increase after approximately 5 CTCs. If distinct prognostic groups are to be identified, women with no CTC could plausibly represent a distinct favorable one.

show abstract

Inflation of the type I error rate when a continuous confounding variable is categorized in logistic regression analyses

Cited by 197 publications

References 24 publications

Adjustment for continuous confounders: an example of how to prevent residual confounding

Adjustment for continuous confounders: an example of how to prevent residual confounding

Predicting the risk of endometrial cancer in postmenopausal women presenting with vaginal bleeding: the Norwich DEFAB risk assessment tool

Modeling the relationship between circulating tumour cells number and prognosis of metastatic breast cancer

Contact Info

Product

Resources

About