Objectives: In-hospital mortality after heart transplant is around 5%. Predicting the risk of in-hospital mortality can be informative for transplant candidacy and prognosis. The Elixhauser Comorbidity Index is an International Statistical Classification of Diseases and Related Health Problems diagnostic code-based comorbidity measurement tool that can predict in-hospital mortality. This study aimed to develop a composite recipient comorbidity and demographic index based on the Elixhauser Comorbidity Index to predict the in-hospital mortality rate of heart transplant recipients. Materials and Methods: This study assessed the in-hospital mortality risk prediction with the Elixhauser Comorbidity Index and demographic variables of heart transplant recipients from the National Inpatient Sample database. A multivariable model that included demographic information and Elixhauser Comorbidity Index was used to assess in-hospital mortality, with Elixhauser Comorbidity Index and age used to develop a single-index adjusted Elixhauser Comorbidity Index. Results: Among 3469 heart transplant patients identified from 2015 (quarter 4) to 2020, in-hospital mortality was 5.13%. Age best predicted (C statistic 0.673; 95% CI, 0.638-0.709) in-hospital mortality, followed by the Elixhauser Comorbidity Index (C statistic 0.638; 95% CI, 0.598-0.678) and race and ethnicity (C statistic 0.571; 95% CI, 0.533-0.609). Sex did not have predictive power (C statistic 0.501; 95% CI, 0.467-0.535). In the multivariable model with demographics, the predictive power of the Elixhauser Comorbidity Index was improved (C statistic 0.753; 95% CI, 0.720-0.785; DeLong P < .001). The single-index adjusted model had comparable discriminative power (C statistic 0.763; 95% CI, 0.731-0.794; DeLong P = .766) to the Elixhauser Comorbidity Index in predicting in-hospital mortality. Both models had good calibration with Brier score <0.05. Conclusions: The Elixhauser Comorbidity Index is an effective measure to predict in-hospital mortality after heart transplant. The improved measure adjusted index could be used as a standardized composite score to account for recipient comorbidity and demographics across clinical studies.

Key words : Clinical outcomes, Elixhauser Comorbidity Index, Prognosis

Introduction

Heart transplant (HT) is an effective therapy to extend the lives of patients with advanced heart disease.¹ As a high-risk procedure, the post-HT in-hospital period has a markedly high mortality of around 5%, compared with 0.7% of all surgical procedures.^1-4 Several risk factors, including age, race and ethnicity, renal failure, and liver function, have been identified.^1,3,4 Thus, risk prediction of in-hospital mortality after HT is informative for transplant candidacy and post-HT prognosis.⁴

The assessment of clinical outcomes often requires risk adjustment for comorbidities.^5,6 The Elixhauser Comorbidity measure was developed to identify a comprehensive set of comorbidities in large-scale administrative data based on the International Classification of Diseases, Ninth/Tenth Revision, Clinical Modification (ICD-9/10-CM) codes.⁷ There are 29 and 38 comorbidities identified in the ICD-9-CM and ICD-10-CM systems, respectively.⁸ Several studies have shown the effectiveness of using Elixhauser Comorbidities to identify the risk of in-hospital mortality.^9-11 Recently, the Elixhauser Comorbidity measure was further developed into the Elixhauser Comorbidity Index (ECI), a weighted single scoring system, to predict in-hospital mortality.¹²

The ECI has shown overall high predictive power (C statistic 0.777; 95% CI, 0.776-0.778) for in-hospital mortality in all subjects during the validation phase of its development.¹² Although the ECI has been validated in several disease categories with top mortality, such as septicemia and pneumonia, in-hospital mortality in HT patients has not been specifically assessed by the ECI.¹²

This study aimed to assess the ECT for in-hospital mortality risk prediction in HT patients who were part of the National (Nationwide) Inpatient Sample (NIS) database, the largest in-patient public database in the United States, accounting for 20% of all hospital discharges.¹³ Because the ECI does not include demographic information, in-hospital mortality risk predictions by demographics, including age, race and ethnicity, and sex, were also assessed. Finally, this study aimed to develop a composite single index, the adjusted ECI (aECI), which could account for all comorbidity and demographic information.The aECI could be used to standardize recipient comorbidity and demographic measurements for clinical HT studies. This is especially useful for studies with small samples, such as single-center data sources. In addition, the standardization of comorbidity and demographic measurement could make it easier to replicate and compare results across different studies.

Materials and Methods

The NIS database was used to identify patients who received HTs based on the ICD-10 Procedure Coding System (02YA0Z0) between the last quarter of 2015 and 2020. Patient demographic data, including age, race and ethnicity, sex, in-hospital mortality record, and ICD-10-CM, were extracted from the database. The Elixhauser Comorbidity software was used to identify 38 Elixhauser Comorbidity and then calculate ECI.⁸ The weights for the Elixhauser Comorbidities are listed in Table 1.

Logistic regression was performed for a best-fit model to predict in-hospital mortality by ECI, age, race and ethnicity, and sex, respectively. Multiva-riable logistic regression was then used to examine the best-fit prediction model for in-hospital mortality by the ECI with demographics (ECID) model, which included ECI, age, race and ethnicity, and sex. The receiver operating characteristic curve for each regression model was plotted. The area under the receiver operating characteristic curve (AUC) was estimated with a 95% confidence interval (95% CI) for each model.

The ability of each model to discriminate in-hospital mortality was characterized with the use of C statistics. The AUC and C statistics have the same numeric values. A C statistic of 0.5 represents that the model has no predictive power, and a C statistic of 1 means the model accurately classifies all in-hospital deaths. Generally, a C statistic of >0.7 indicates the presence of good predictive variables. The 2-tailed DeLong test was used to compare 2 different AUCs.¹⁴ The AUC evaluates only the model’s capacity to distinguish between the presence of mortality. In this scenario, evaluation of calibration is also crucial because accurate discrimination does not necessarily indicate good calibration. There may be instances where good discrimination is associated with low calibration, which may result in under or over prediction of mortality.¹⁵ The Brier score was used to measure the level of calibration, which assesses the difference between the anticipated and observed risk.¹⁶ A lower Brier score indicates that the predicted and actual risks are more closely aligned, indicating that the index is better calibrated.¹⁶

Because race/ethnicity and sex had small pre-dictive powers for in-hospital mortality, ECI and age were chosen to develop the single-index aECI based on the Sullivan method.^17,18 The parameter estimates of ECI and age from the ECID multivariable regres-sion were extracted. Weights should be calculated by dividing all regression coefficients by the smallest regression coefficient and then rounding the quotient to the nearest integer. This way, a larger weight indicates a stronger association by proportion to the variable with the smallest coefficient. In this case, however, there were only 2 variables (ie, ECI and age) and the quotient of their regression coefficients was around 1.5; rounding to the nearest integer would introduce too much error. Instead, weights were calculated by dividing the regression coefficients by half of the smallest coefficient, which was followed by rounding the quotient to the nearest integer. A weighted addition of ECI and age was used to calculate a single index of aECI.

The study was retrospective based on the open NIS database and was exempted from Institutional Review Board approval. The study protocols conformed to the ethical guidelines of the 1975 Helsinki Declaration. All analyses were performed with SAS version 9.4.

Results

From the NIS database records of the last quarter of 2015 through 2020, 3469 HT patients were identified. The Elixhauser Comorbidities for this cohort are summarized in Table 1. The most common comor-bidities (>10% of patients) were complicated hyper-tension (1153; 33.24%), diabetes with chronic complications (907; 26.15%), heart failure (645; 18.59%), hypothyroidism (434; 12.51%), obesity (413; 11.91%), and chronic pulmonary disease (403; 11.62%).

The demographics and characteristics of patients are summarized in Table 2. The mean age of the cohort was 47.79 ± 19.79 years. Age of HT recipients showed an overall left-skewed distribution and a spike in infant HTs. Race and ethnicity among the cohort included 1983 White (57.16%), 711 African American (20.50%), 362 Hispanic (10.44%), 132 Asian (3.81%), and 14 Native American (0.40%) HT recipients, in addition to 113 HT recipients of other races/ethnicities. The cohort identified 2462 (70.97%) male and 1007 (29.03%) female HT recipients. Among those in the cohort, 178 in-hospital mortalities were recorded, with an overall mortality rate of 5.13%. The mean ECI was 0.97 ± 7.04 with a minimal score of -30 and a maximal score of 25. The distribution of age and ECI are shown in Figure 1, which showed a roughly normal distribution.

The C statistics for ECI, age, race/ethnicity, and sex are shown in Figure 2 and summarized in Table 3. Age best predicted (C statistic 0.673; 95% CI, 0.638-0.709) in-hospital mortality, followed by ECI (C statistic 0.638; 95% CI, 0.598-0.678) and race/ethnicity (C statistic 0.571; 95% CI, 0.533-0.609). Sex did not have a predictive power (C statistic 0.501; 95% CI, 0.467-0.535) for in-hospital mortality. None of the single parameters had a significant C statistic (over 0.70). Inclusion of ECI, age, race/ethnicity, and sex in a multivariable logistic regression for the ECID model is shown in Figure 3 and summarized in Table 3. The ECID model had significant discriminative power (C statistic 0.753; 95% CI, 0.720-0.785) for in-hospital mortality and was higher than ECI alone (DeLong test P < .001).

The results for the aECI model are shown in Figure 1 and Figure 3 and summarized in Table 2 and Table 3. The weights for ECI and age were 3 and 2, respectively, with aECI calculated as aECI = 2 × age + 3 × ECI. The mean aECI was 98.50 ± 45.74 with a minimal score of -21 and a maximal score of 201. The distribution of aECI is shown in Figure 1, showing a roughly normal distribution. The single index aECI had comparable predictive power (C statistic 0.763; 95% CI, 0.731-0.794; DeLong test P = .766) for in-hospital mortality compared with the multivariable ECID model. Consequently, the aECI had higher discriminative power than ECI alone (DeLong test P < .001). All models had good calibration, with Brier score <0.05.

Discussion

This study examined predicted mortality by ECI and demographics in recent HT patients in the NIS database from the last quarter of 2015 to 2020. Both ECI and age were found to be moderate predictors for in-hospital mortality, whereas race/ethnicity and sex were weaker predictors. A multivariable regression, which included ECI and all 3 demographic variables, increased model prediction to a significant level. Both ECI and age were chosen to develop a new single-index aECI, which showed comparable effectiveness in predicting in-hospital mortality in HT.

The ECI gave a moderate but not significant prediction (C statistic 0.638) of in-hospital mortality post-HT. Thus, comorbidities alone did not have sufficient discriminative power, and the inclusion of additional demographic variables was justified.

Recipient age at transplant was the strongest single predictor for in-hospital mortality in HT patients. This was incongruent with previous studies, including Singh and colleagues, who identified age >65 years as a risk factor (odds ratio = 1.89) in their risk stratification model.⁴ It was proposed that aging compromised the immune function, which led to an increase in all-cause mortality.¹⁹

Race/ethnicity had little effect on post-HT in-hospital mortality, which confirmed the previous results from Singh and colleagues, in which patients in all race and ethnicity groups benefited equally from HT.³ However, Singh and colleagues noted that long-term survival only improved in White patients but not in African American or Hispanic patients.³ Additional modifiers of race and ethnicity might need to be included in the aECI formula if future studies explore the long-term prediction of mortality by ECI.

Sex did not affect the prediction of in-hospital mortality in HT, which was consistent with previous studies.^20,21 Sex mismatch (eg, female heart in male recipient), however, has been widely shown to have a worse prognosis.^22-25 The effect of sex mismatch can be examined in future studies if donor information is recorded.

With additional demographic information included, the multivariable ECID offered a much better prediction of mortality. The further simpli-fication of ECID with only ECI and age into a single aECI score maintained the high predictive power. The single-index aECI (C statistic 0.763) could better discriminate the presence of mortality than the Singh model (C statistic 7.22), where age, specific diagnosis, mechanical support type, ventilator support, esti-mated glomerular filtration rate, and total serum bilirubin were used in the model.⁴

The ECI calculations are free and readily accessible. For clinicians and patients, ECI can be calculated by a free online ECI calculator based on past medical history.²⁶ For researchers and health care administrators, ECI can be calculated by the Elixhauser Comorbidity Software from NIS.⁸ Given that aECI can be easily derived from ECI, aECI has the advantage of being an easy reference for post-HT in-hospital mortality. This is especially useful for health care administration, such as the NIS database, where only diagnosis codes and patient demographic information are recorded. More importantly, aECI can be used as a standardized measurement of recipient comorbidity and demographic information in HT clinical studies. When the presence of comorbidities in the sample is low, especially among single-center data sources, comorbidities cannot be well-adjusted using multivariable analysis. Thus, standardization of comorbidity and demographic measurements by the aECI can help replicate and compare results across studies.

This study had several limitations. First, all mortality examined was in-hospital without any follow-up. It was argued that, to examine early posttransplant outcomes, a 30-day perioperative period should be followed.²⁷ Second, Elixhauser Comorbidities were recorded as binary (present/not present) so that, within the same comorbidity, patients were more heterogenous with different extents/staging of the disease. Third, donor infor-mation has a well-known effect on survival in HT patients. However, NIS only records recipient information. Relevant donor factors, such as donor age, sex, and ischemic time, are not recorded in the NIS database. Thus, only a composite recipient comorbidity and demographic index can be developed. Donor information needs to be accounted for separately in future studies.

Conclusions

This study showed ECI was an effective measure to predict post-HT in-hospital mortality. The improved measure aECI can be used as a single index to standardize the characterization of recipient comor-bidity and demographic information across clinical studies.

References:

1. Stehlik J, Edwards LB, Kucheryavaya AY, et al. The Registry of the International Society for Heart and Lung Transplantation: twenty-seventh official adult heart transplant report--2010. J Heart Lung Transplant. 2010;29(10):1089-1103. doi:10.1016/j.healun.2010.08.007
   CrossRef - PubMed
   
2. Heeney A, Hand F, Bates J, Mc Cormack O, Mealy K. Surgical mortality - an analysis of all deaths within a general surgical department. Surgeon. 2014;12(3):121-128. doi:10.1016/j.surge.2013.07.005
   CrossRef - PubMed
   
3. Singh TP, Almond C, Givertz MM, Piercey G, Gauvreau K. Improved survival in heart transplant recipients in the United States: racial differences in era effect. Circ Heart Fail. 2011;4(2):153-160. doi:10.1161/CIRCHEARTFAILURE.110.957829
   CrossRef - PubMed
   
4. Singh TP, Almond CS, Semigran MJ, Piercey G, Gauvreau K. Risk Prediction for early in-hospital mortality following heart transplantation in the United States. Circ Heart Fail. 2012;5(2):259-266. doi:10.1161/CIRCHEARTFAILURE.111.965996
   CrossRef - PubMed
   
5. Harrison C, Britt H, Miller G, Henderson J. Examining different measures of multimorbidity, using a large prospective cross-sectional study in Australian general practice. BMJ Open. 2014;4(7):e004694. doi:10.1136/bmjopen-2013-004694
   CrossRef - PubMed
   
6. Pefoyo AJK, Bronskill SE, Gruneir A, et al. The increasing burden and complexity of multimorbidity. BMC Public Health. 2015;15:415. doi:10.1186/s12889-015-1733-2
   CrossRef - PubMed
   
7. Elixhauser A, Steiner C, Harris DR, Coffey RM. Comorbidity measures for use with administrative data. Med Care. 1998;36(1):8-27. doi:10.1097/00005650-199801000-00004
   CrossRef - PubMed
   
8. Elixhauser Comorbidity Software Refined for ICD-10-CM. Accessed March 25, 2023. https://hcup-us.ahrq.gov/toolssoftware/comorbidityicd10/comorbidity_icd10.jsp
   CrossRef - PubMed
   
9. Menendez ME, Neuhaus V, van Dijk CN, Ring D. The Elixhauser Comorbidity Method outperforms the Charlson Index in predicting inpatient death after orthopaedic surgery. Clin Orthop Relat Res. 2014;472(9):2878-2886. doi:10.1007/s11999-014-3686-7
   CrossRef - PubMed
   
10. Metcalfe D, Masters J, Delmestri A, et al. Coding algorithms for defining Charlson and Elixhauser co-morbidities in Read-coded databases. BMC Med Res Methodol. 2019;19(1):115. doi:10.1186/s12874-019-0753-5
    CrossRef - PubMed
    
11. Stukenborg GJ, Wagner DP, Connors AF. Comparison of the performance of two comorbidity measures, with and without information from prior hospitalizations. Med Care. 2001;39(7):727-739. doi:10.1097/00005650-200107000-00009
    CrossRef - PubMed
    
12. Moore BJ, White S, Washington R, Coenen N, Elixhauser A. Identifying Increased Risk of Readmission and In-hospital Mortality Using Hospital Administrative Data: The AHRQ Elixhauser Comorbidity Index. Medical Care. 2017;55(7):698-705. doi:10.1097/MLR.0000000000000735
    CrossRef - PubMed
    
13. NIS Database Documentation. Accessed March 25, 2023. https://hcup-us.ahrq.gov/db/nation/nis/nisdbdocumentation.jsp.
    CrossRef - PubMed
    
14. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44(3):837-845.
    CrossRef - PubMed
    
15. Alba AC, Agoritsas T, Walsh M, et al. Discrimination and calibration of clinical prediction models: users’ guides to the medical literature. JAMA. 2017;318(14):1377-1384. doi:10.1001/jama.2017.12126
    CrossRef - PubMed
    
16. Bae S, Massie AB, Caffo BS, Jackson KR, Segev DL. Machine learning to predict transplant outcomes: helpful or hype? Transpl Int. 2020;33(11):1472-1480. doi:10.1111/tri.13695
    CrossRef - PubMed
    
17. Sullivan LM, Massaro JM, D’Agostino RB. Presentation of multivariate data for clinical use: the Framingham Study risk score functions. Stat Med. 2004;23(10):1631-1660. doi:10.1002/sim.1742
    CrossRef - PubMed
    
18. van Walraven C, Austin PC, Jennings A, Quan H, Forster AJ. A modification of the Elixhauser comorbidity measures into a point system for hospital death using administrative data. Med Care. 2009;47(6):626-633. doi:10.1097/MLR.0b013e31819432e5
    CrossRef - PubMed
    
19. Wever-Pinzon O, Edwards LB, Taylor DO, et al. Association of recipient age and causes of heart transplant mortality: implications for personalization of post-transplant management-An analysis of the International Society for Heart and Lung Transplantation Registry. J Heart Lung Transplant. 2017;36(4):407-417. doi:10.1016/j.healun.2016.08.008
    CrossRef - PubMed
    
20. De Santo LS, Marra C, De Feo M, Amarelli C, Romano G, Cotrufo M. The impact of gender on heart transplantation outcomes: a single center experience. Ital Heart J. 2002;3(7):419-423.
    CrossRef - PubMed
    
21. Moayedi Y, Fan CPS, Cherikh WS, et al. Survival outcomes after heart transplantation. Circ Heart Fail. 2019;12(10):e006218. doi:10.1161/CIRCHEARTFAILURE.119.006218
    CrossRef - PubMed
    
22. Al-Khaldi A, Oyer PE, Robbins RC. Outcome analysis of donor gender in heart transplantation. J Heart Lung Transplant. 2006;25(4):461-468. doi:10.1016/j.healun.2005.11.456
    CrossRef - PubMed
    
23. Ayesta A. Influence of sex-mismatch on prognosis after heart transplantation. Front Cardiovasc Med. 2021;8:617062. doi:10.3389/fcvm.2021.617062
    CrossRef - PubMed
    
24. Doulamis IP, Tzani A, Kourek C, et al. Sex mismatch following heart transplantation in the United States: Characteristics and impact on outcomes. Clin Transplant. 2022;36(12):e14804. doi:10.1111/ctr.14804
    CrossRef - PubMed
    
25. Reed RM, Netzer G, Hunsicker L, et al. Cardiac size and sex matching in heart transplantation. JACC Heart Fail. 2014;2(1):73-83. doi:10.1016/j.jchf.2013.09.005
    CrossRef - PubMed
    
26. Free Online Elixhauser Comorbidity Index Calculator - OrthoToolKit. Accessed March 25, 2023. https://orthotoolkit.com/elixhauser-comorbidity-index/
    CrossRef - PubMed
    
27. Volpp KG, Rosen AK, Rosenbaum PR, et al. Mortality among hospitalized Medicare beneficiaries in the first 2 years following ACGME resident duty hour reform. JAMA. 2007;298(9):975-983. doi:10.1001/jama.298.9.975
    CrossRef - PubMed