Original Article
Categorizing BMI may lead to biased results in studies investigating in-hospital mortality after isolated CABG

https://doi.org/10.1016/j.jclinepi.2007.01.008Get rights and content

Abstract

Objective

To investigate how categorizing body mass index (BMI) into weight classes can impact the assessment of the relationship between BMI and in-hospital mortality after coronary artery bypass graft (CABG) surgery.

Study Design and Setting

BMI–mortality (in-hospital) relationship was assessed in 5,762 patients who underwent isolated CABG at Baylor University Medical Center (Dallas, TX) from January 1, 1997 to November 30, 2003. Different ways of modeling BMI were used to investigate this association in a propensity-adjusted model, controlling for risk factors identified by the Society of Thoracic Surgeons (STS) and other clinical/nonclinical details.

Results

A highly significant (P = 0.003) association between BMI (modeled with a restricted cubic spline) and mortality was found. Reduced risk of in-hospital mortality was observed for subjects with BMI in the low-30s as compared with patients with BMI in the mid-20s or over 40 kg/m2. Results were strongly affected by the way BMI was specified in the multivariable model. Only five of the 10 BMI categorizations considered produced significant results, and these results did not fully determine the effect of BMI on mortality.

Conclusions

BMI categorization critically impacts study results. Conceivably, findings of other studies investigating BMI and adverse outcomes after CABG may be similarly affected. Investigators should consider smoothing techniques to avoid categorization.

Introduction

With the rising incidence of obesity in the U.S. population, there has been increasing interest in how obesity affects the risk of disease as well as the risk of adverse events after surgery or other treatment. Upon reviewing the literature, we found great variation in the categorizations of body mass index (BMI) used in studies examining the effect of BMI on risk of postoperative adverse outcomes. It is possible that this variation in how subjects are grouped by BMI compromises the internal validity of some studies assessing the risks associated with obesity, and may lead researchers to inaccurate conclusions. It has previously been suggested that the variation in the categorization of BMI may be partly responsible for the variable and often conflicting findings regarding the effect of BMI on risk for adverse events after coronary artery bypass graft (CABG) surgery [1]. Specifically, the way BMI is specified in the statistical model used for the analysis (i.e., BMI categorization schemes as well as assuming a linear effect of BMI on the outcome of interest) may critically affect the internal validity of epidemiologic studies focusing on the relationship between obesity and adverse outcomes after cardiac surgery. Although the significant problems associated with categorizing continuous exposures/covariates have been noted in the statistical and epidemiologic literature [2], [3], [4], such categorizations remain commonplace in medical research and, in particular, research investigating adverse outcomes after CABG. To our knowledge, no attempt to determine the extent to which BMI categorization (or assuming a linear relationship between BMI and the post-CABG adverse outcome) may influence study results has been made. Presented here is a study investigating the relationship between BMI and risk of in-hospital mortality after CABG using categorization schemes and analytic methods used in other investigations focusing on this topic. We furthermore suggest a way of modeling the effect of BMI that does not require categorization or an assumption of a linear effect.

Cubic spline functions can be used to examine the strength and shape of relationships between continuous variables and the outcome of interest [3], [5], [6]. These functions permit departures from linearity of the associations between outcomes and continuous exposures/covariates and provide vital information for the development of the final multivariable model. Most importantly, use of these functions avoids the need for categorization of continuous exposures and covariates, which is inefficient and often arbitrary [7]. Unfortunately, despite the ready availability of statistical software capable of handling techniques such as nonparametric and spline regression, these tools are seldom used and epidemiologic studies usually present estimates from categorical analysis in place of smoothed assessment of continuous variables/exposures [7]. Likewise, the vast majority of studies investigating the relationship between BMI and mortality during/after CABG [1], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19] used predefined BMI categorizations, for example, World Health Organization, American Heart Association, or arbitrary BMI categorization based on the distribution of BMI in a single patient cohort. If continuous variables are categorized, the cutoffs should be dependent on the shape of the association between these variables and the outcome, and not on the distribution of the variables alone because the decision on where to establish the cutoff values for categorizing continuous exposures (and/or covariates) can critically affect the internal validity of the investigation [2], [4] and may have major implications for public health.

In addition to the variety of categorization schemes applied to BMI in the literature addressing the association between BMI and mortality after CABG, a handful of studies considered BMI as a linear effect or used a transformation of BMI as a continuous predictor in the model [18], [20], [21]. However, no study has based its conclusions on a smoothed function of BMI. Although smoothing methods, such as cubic spline functions, are among the most efficient techniques for determining appropriate clinical groupings of continuous variables, the optimal utilization of such methods in epidemiologic studies may be to estimate the strength and direction of the relationships between continuous variables and outcomes when these associations show possible departures from linearity. The use of cubic spline functions eliminates the possibility of mistakenly assuming a linear relationship between the predictors and the outcome of interest (in this case, the logit of the probability of in-hospital mortality) and provides a more efficient method of treating continuous variables. Cubic splining obviates the need for categorizing the continuous variables altogether.

The present study examines the relationship between BMI and risk of in-hospital mortality after CABG in a cohort of 5,762 patients who underwent CABG at Baylor University Medical Center (Dallas, TX) between January 1, 1997 and November 30, 2003. BMI is investigated with the use of a restricted cubic spline function. Findings are then contrasted with those produced by analyses, conducted for the same cohort of patients, modeling BMI according to analytic methods identified in the literature from studies investigating this same subject.

Section snippets

BMI and in-hospital mortality literature

We conducted a PubMed search to identify studies that investigated the effect of BMI on in-hospital mortality after CABG. Search strategies were formulated to retrieve records published in English that combined terms related to BMI, CABG, and operative or in-hospital mortality. Reference lists of review articles were searched for additional references. From the resulting list of articles, we selected studies that investigated the effect of BMI on mortality after isolated CABG, reviewed the

Results

Patient characteristics including clinical and surgical details and the distribution of BMI are presented in Table 2 and Fig. 2, respectively. BMI ranged from 13.0 to 66.1 kg/m2 and the mean (SD) for BMI was 28.6 (5.6) kg/m2 for patients included in the study cohort.

In-hospital death occurred in 176 (3.1%) subjects included in the study cohort. The propensity-adjusted model for the association between BMI (modeled with a restricted cubic spline function) and mortality was highly significant

Discussion

When the association between BMI and in-hospital mortality was estimated using BMI as a restricted cubic spline function, BMI was found to be a strong predictor of mortality (see Fig. 3 for the shape of the association). However, when BMI was analyzed under the various categorization schemes identified in the literature, findings were inconsistent, demonstrating that categorization can greatly influence study results. Thus, it appears that the way BMI is specified in a multivariable model plays

Acknowledgments

The authors thank Briget da Graca for writing and editorial assistance in preparing this article.

References (26)

  • D.S. Lee et al.

    Predicting mortality among patients hospitalized for heart failure: derivation and validation of a clinical model

    JAMA

    (2003)
  • S.M. O'Brien

    Cutpoint selection for categorizing a continuous predictor

    Biometrics

    (2004)
  • G. Prabhakar et al.

    The risks of moderate and extreme obesity for coronary artery bypass grafting outcomes: a study from the Society of Thoracic Surgeons' database

    Ann Thorac Surg

    (2002)
  • Cited by (75)

    • Postoperative atrial fibrillation: Sex-specific characteristics and effect on survival

      2020, Journal of Thoracic and Cardiovascular Surgery
      Citation Excerpt :

      The propensity score was estimated by modeling the risk factors reported in Table 1 (recognized by the STS as affecting patient outcomes17) as independent variables in a multivariable logistic regression model with sex as the outcome. Continuous variables were modeled with restricted cubic splines (5 knots).18,21 The propensity score (also modeled with 5-knot restricted cubic spline functions) was as an independent variable in:

    • Underestimation of the incidence of new-onset post–coronary artery bypass grafting atrial fibrillation and its impact on 30-day mortality

      2017, Journal of Thoracic and Cardiovascular Surgery
      Citation Excerpt :

      The impact of limiting the definition of post-CABG AF to events requiring treatment on the detection of new-onset AF in the US population undergoing isolated CABG was estimated by applying the risk-adjusted difference in incidence between patients whose AF events were versus were not identified in the STS data in our study population to the total number of isolated CABG procedures performed in the United States (reported for the most recent full year in the STS Adult Cardiac Surgery database).24 To determine the effect of the definition of post-CABG AF used (identified in the STS data vs not) on 30-day mortality, we used a GEE logistic model with 30-day mortality as the outcome variable and post-CABG AF (no AF, AF according to STS data, and AF not identified in the STS data) as the main predictor, with model covariates including the STS risk of mortality21 and time of operation, modeled using the approach and spline functions22,23 described earlier. Model estimates were used to compute ORs, 95% CIs, and P values.

    View all citing articles on Scopus
    View full text