Begin typing your search above and press return to search.
Volume: 20 Issue: 9 September 2022

FULL TEXT

ARTICLE
Evaluation of a Flexible Weight and Height Interaction Versus Body Mass Index for Mortality Prediction After Adult Heart Transplant

Abstract

Objectives: Body mass index (calculated as kilograms body weight divided by the square of height in meters) is a known predictor of mortality after adult heart transplant but has limitations. We investigated whether inclusion of an explicit weight-height interaction effect improves prediction of mortality after heart transplant compared with body mass index.
Methods and Results: We included a cohort of 46 424 adults who had undergone heart transplant as documented in the United Network for Organ Sharing database. Risk-adjusted prediction models for 1-year and 5-year mortality were constructed, one with the flexible weight-height interaction and the other with the body mass index. Overall model performance (R2) and discrimination (the Harrell concordance probability C index and the Somers Dxy rank correlation) were calculated. Compared with the body mass index model, the weight-height model had slightly improved overall performance (R2, 0.316 vs 0.313) and 1-year mortality discrimination (optimism-corrected Harrell C, 0.642 vs 0.640; Somers Dxy, 0.284 vs 0.281). Compared with the body mass index model, the weight-height model had improved overall performance (R2, 0.232 vs 0.224) and similar discrimination (optimism-corrected Harrell C, 0.600 vs 0.599; Somers Dxy, 0.200 vs 0.197) for 5-year mortality.
Conclusions: Allowance for a flexible relationship between height and weight did not appreciably improve mortality prediction after heart transplant, versus body mass index, although additional research is warranted.


Key words : Cardiac transplantation, Obesity, Risk prediction

Introduction

Extreme presurgical body mass index (BMI, calculated as kilograms body weight divided by the square of height in meters) in individuals undergoing heart transplant (HT) is associated with higher posttransplant mortality.1 More specifically, a BMI under 18.5 or over 30 is associated with the greatest risk. In fact, guidance from the International Society for Heart and Lung Transplantation suggests that individuals should take steps to reduce their BMI below 30 prior to being listed for HT.2 However, studies have identified limitations and potential biases to BMI as a measure of adiposity.3 The BMI assumes the relationship between weight and height to be fixed. That means, for a given weight, the mortality risk after HT could differ based on an individual’s height. Thus, the inflexible metric of BMI may not be optimal for utilizing the weight-height relationship to predict outcomes. Recent studies have suggested that alternative relationships between weight and height may better predict cardiovascular outcomes compared with BMI.4,5 Whether these findings extend to outcomes after HT is unclear. This study investigated whether inclusion of an explicit weight-height interaction effect improves prediction of mortality after HT compared with BMI.

Materials and Methods

This study used data from the United Network for Organ Sharing database, collected by the Organ Procurement and Transplantation Network, current through January 2022. The database contains deidentified patient-level records including recipients, donors, and candidates on the wait list. As the registry provided a dataset without patient or center identifiers, our study did not require Institutional Review Board approval.

This retrospective cohort study included adults (aged ≥18 years) who underwent HT from January 1, 2000, to December 31, 2021. We excluded individuals who were <18 years of age at transplant, transplants occurring prior to January 1, 2000, and those who underwent multiorgan transplants. The final cohort consisted of 46 424 HT recipients.

The outcomes of interest included the time to mortality within the first year and the first 5 years after successful HT. We also collected information about the overall cohort, including key recipient, donor, and procedural variables of interest (Table 1). To account for potential clerical errors in the database, recipient height was limited to a range of 90 to 230 cm and weight was limited to a range of 35 to 250 kg (otherwise, each was assigned a status of “missing”). Recipient-donor sex match was categorized as a female recipient matched with a female donor, male recipient matched with a male donor, female recipient matched with a male donor, and male recipient matched with a female donor.

We imputed missing data using multiple impu-tations with chained equations, with “missingness” assumed to be random, via predictive mean-matching with the “mice” package (version 3.13.0) in R.6 All predictors and the outcomes were included in the imputation model. We generated 20 multiply imputed datasets with 20 iterations. Outcomes were analyzed with a Cox proportional hazards model using the “rms” package (version 6.2-0) in R on each imputed dataset and pooled using the Rubin rules. For each analysis, we constructed 2 models: (1) a maximally flexible model representing height and weight (each log-transformed) as separate, nonlinear (restricted cubic spline with 3 knots) terms and allowing for interactions; and (2) a log-transformed BMI model. We adjusted each model for heart failure etiology, recipient-donor sex match, recipient age, history of diabetes, ventilator use at transplant, pulmonary vascular resistance and kidney function at transplant, extracorporeal membrane oxygenation use at transplant, donor age, and transplant era (2000-2005 [referent], 2006-2010, 2011-2017, and 2018-2021).7 The models for 5-year mortality were adjusted for heart failure etiology, recipient-donor sex match, recipient age, history of diabetes, pulmonary vascular resistance, kidney function at transplant, and donor age.

Model performance was assessed by evaluating calibration and discrimination measures and compared between models using the maximally flexible height and weight interaction and BMI. Calibration, which assessed agreement between predicted and observed survival, was estimated by a bootstrap (200 samples) internal validation procedure. Calibration plots were constructed. Discrimination, which assesses how well the constructed prediction model appropriately identifies those with and without the event, was evaluated using the Harrell concordance statistic (C index), the Somers Dxy rank correlation (index of discrimination between predicted score and observed responses), and the optimism-adjusted R2 (as a measure of overall model fit) by a bootstrap (200 samples) procedure. Event-free probabilities at 1 year and 5 years were estimated across a range of weight-height combinations (weight range = 50-200 kg by kg; height range = 160-200 cm by 5 cm). We performed data management with SAS (version 9.4; SAS Institute) and data analysis with R (version 4.0.4; The R Foundation, https://www.r-project.org/).

Results

Characteristics of the 46 424 HT recipients are shown in (Table 1). The median (25th, 75th percentile) age was 56 years (46, 62), most recipients were White (68.1%) and were men (74.3%), had a dilated cardiom-yopathy heart failure pathogenesis (48.2%), had a median BMI of 26.9 (23.7, 30.5), had a median donor age of 30 years (22, 41), and had a median duration of follow-up of 4.9 years (1.7, 9.6).

Within our included population, the baseline survival at 1 year was 89.9%. The maximally flexible weight-height models had marginally better performance for 1-year mortality (Figure 1A) compared with the BMI model (Figure 1B). Compared with the BMI model, the weight-height model had slightly improved overall performance (R2, 0.316 vs 0.313) and discrimination (optimism-corrected Harrell C, 0.642 vs 0.640; Somers Dxy, 0.284 vs 0.281). The flexible weight-height model predicted higher 1-year mortality (12.7% vs 11.2%) compared with the BMI model. Increasing recipient weight, decreasing recipient height, and increasing recipient BMI were associated with higher 5-year mortality (Figure 2).

The baseline survival at 5 years was 80.5%. The maximally flexible weight-height models had similar performance for 5-year mortality (Figure 1C) compared with the BMI model (Figure 1D). Compared with the BMI model, the weight-height model had improved overall performance (R2, 0.232 vs 0.224) and similar discrimination (optimism-corrected Harrell C, 0.600 vs 0.599; Somers Dxy, 0.200 vs 0.197). The flexible weight-height model predicted higher 5-year mortality (22.5% vs 18.5%) compared with the BMI model. Increasing recipient weight, decreasing recipient height, and increasing recipient BMI were associated with higher 5-year mortality (Figure 2).

Discussion

This evaluation of a modern cohort of adults who underwent HT suggests that use of weight and height as independent and interacting variables improves performance of models predicting mortality compared with BMI, when adjusted for known confounders. This is evidenced by the improved overall model performance (higher R2) value and model discrimination (higher Harrell C and Somers Dxy). However, the differences were marginal for 1-year mortality and similar for 5-year mortality. The most notable difference is in the model’s mortality prediction (when holding other variables at their median value), which was higher for the flexible weight-height model compared with BMI. These findings are somewhat in opposition to studies by Shuey and colleagues5 and Sorjonen and colleagues,4 who both suggested that the use of weight and height improves prediction of mortality and other outcomes in broader cardiovascular populations, rather than the use of BMI.

Despite evidence of associations between BMI and worsened outcomes after HT,1 as well as development of cardiometabolic disease more broadly,8 the use of BMI has known limitations. The BMI calculation assumes a fixed relationship between height and weight across the risk spectrum. However, Shuey and colleagues5 showed that coronary artery disease prediction was reduced in individuals with height over 180 cm and weight under 50 kg; this observation was not seen in the BMI model. Outside of cardiology, greater height has been associated with higher risk of cancer mortality,9 whereas low body weight has been associated with greater suicide risk.10 These relationships may be confounded if evaluated with the BMI calculation. In fact, studies have shown that the weight-to-height ratio in the BMI calculation correlates differently in women and older adults compared with men.11 The flexible relationship between height and weight used in our study predicted higher mortality compared with BMI but did not appreciably improve model performance or discrimination. When each factor was modeled separately as a continuous variable, weight and BMI had similar relationships with posttransplant mortality, increasing risk as their values rose (most notably at the higher range), whereas mortality risk was highest with shorter heights. This suggests that mortality risk for a given BMI may differ based on the underlying weight and height values; this relationship deserves additional research.

Some limitations to this study deserve mention. Although we included factors known to be associated with 1-year and 5-year mortality rates in HT individuals, the influence of other variables or unmeasured variables on outcomes is unknown. We were also unable to compare our model against models containing other measures of adiposity, such as waist circumference, as these characteristics are not available in the United Network for Organ Sharing database. The problem of absent data is also a concern with registry information. Rather than using methods such as multiple imputation, we chose to limit the population (given its large size and the frequency of the outcome) by using only the records that contained complete data.

We showed that allowance for a flexible relationship between height and weight, as opposed to forcing the relationship through BMI, did not appreciably improve mortality prediction following HT. Additional research is needed to better identify the optimal method for using weight and height to improve the care of adults undergoing HT.


References:

  1. Foroutan F, Doumouras BS, Ross H, Alba AC. Impact of pretransplant recipient body mass index on post heart transplant mortality: a systematic review and meta-analysis. Clin Transplant. 2018;32(8):e13348. doi:10.1111/ctr.13348
    CrossRef - PubMed
  2. Mehra MR, Canter CE, Hannan MM, et al. The 2016 International Society for Heart Lung Transplantation listing criteria for heart transplantation: a 10-year update. J Heart Lung Transplant. 2016;35(1):1-23. doi:10.1016/j.healun.2015.10.023
    CrossRef - PubMed
  3. Jackson AS, Stanforth PR, Gagnon J, et al. The effect of sex, age and race on estimating percentage body fat from body mass index: the Heritage Family Study. Int J Obes Relat Metab Disord. 2002;26(6):789-796. doi:10.1038/sj.ijo.0802006
    CrossRef - PubMed
  4. Sorjonen K, Nilsonne G, Falkstedt D, Hemmingsson T, Melin B, Ingre M. A comparison of models with weight, height, and BMI as predictors of mortality. Obes Sci Pract. 2021;7(2):168-175. doi:10.1002/osp4.473
    CrossRef - PubMed
  5. Shuey MM, Huang S, Levinson RT, et al. Exploration of an alternative to body mass index to characterize the relationship between height and weight for prediction of metabolic phenotypes and cardiovascular outcomes. Obes Sci Pract. 2022;8(1):124-130. doi:10.1002/osp4.543
    CrossRef - PubMed
  6. van Buuren S, Groothuis-Oudshoorn K. Mice: multivariate imputation by chained equations in R. J Stat Software. 2011;45(3):1-67. doi:10.18637/jss.v045.i03
    CrossRef - PubMed
  7. Chambers DC, Perch M, Zuckermann A, et al. The International Thoracic Organ Transplant Registry of the International Society for Heart and Lung Transplantation: Thirty-eighth adult lung transplantation report 2021; Focus on recipient characteristics. J Heart Lung Transplant. 2021;40(10):1060-1072. doi:10.1016/j.healun.2021.07.021
    CrossRef - PubMed
  8. Lyall DM, Celis-Morales C, Ward J, et al. Association of body mass index with cardiometabolic disease in the UK Biobank: a Mendelian randomization study. JAMA Cardiol. 2017;2(8):882-889. doi:10.1001/jamacardio.2016.5804
    CrossRef - PubMed
  9. Wiren S, Haggstrom C, Ulmer H, et al. Pooled cohort study on height and risk of cancer and cancer death. Cancer Causes Control. 2014;25(2):151-159. doi:10.1007/s10552-013-0317-7
    CrossRef - PubMed
  10. Amiri S, Behnezhad S. Body mass index and risk of suicide: a systematic review and meta-analysis. J Affect Disord. 2018;238:615-625. doi:10.1016/j.jad.2018.05.028
    CrossRef - PubMed
  11. Sperrin M, Marshall AD, Higgins V, Renehan AG, Buchan IE. Body mass index relates weight to height differently in women and older adults: serial cross-sectional surveys in England (1992-2011). J Public Health (Oxf). 2016;38(3):607-613. doi:10.1093/pubmed/fdv067
    CrossRef - PubMed


Volume : 20
Issue : 9
Pages : 849 - 853
DOI : 10.6002/ect.2022.0182


PDF VIEW [818] KB.
FULL PDF VIEW

From the 1University of Connecticut School of Pharmacy, Department of Pharmacy Practice, Storrs; and the 2Hartford HealthCare Heart and Vascular Institute, Hartford Hospital, Hartford, Connecticut, USA
Acknowledgements: The authors have not received any funding or grants in support of the presented research or for the preparation of this work and have no declarations of potential conflicts of interest.
Corresponding author: William L. Baker, University of Connecticut School of Pharmacy, 69 N Eagleville Rd, Storrs, CT 06269, USA
Phone: +1 860 972 7918
E-mail: william.baker_jr@uconn.edu