Mortality of NBA Players: Risk Factors and Comparison with the General US Population

Featured Application: This is one of the ﬁrst studies using such an extensive cohort of professional basketball players and Cox proportional hazards models, which takes into account left-truncation, i.e., late entry at the age of the players’ NBA debuts. Results suggested that height and ethnicity are associated to mortality: White players and small players live longer. Our study is useful for devising strategies for health interventions and the proper allocation of resources with respect to the general population. Abstract: Concerns have been raised recently by players’ associations regarding the risk of death among retired players. Using a retrospective cohort study, we analyzed factors associated with the mortality of National Basketball Association (NBA) players and compared their life expectancy with that of the general population. We analyzed a cohort of 3985 players who participated in the NBA from its inception in 1946 to April 2015 (481 active and 3504 former players). We used the data for the 3504 former NBA players, of whom 687 (19.1%) died before 15 April 2015, to study the elapsed time between the end of their NBA careers until death. Cox proportional hazards models were employed in the multivariate survival analysis. After adjusting for age at the end of the NBA career and calendar year, we found that mortality is associated with height and ethnicity. Taller players and African-American players had a higher instantaneous risk of death than shorter players or white players. In addition, the life expectancy of players (regardless of height and ethnicity) has increased since the inception of the NBA. This is one of the ﬁrst studies using such an extensive cohort of professional basketball players and Cox proportional hazards models. Results conﬁrmed that height is associated with mortality. In addition, ethnicity is also linked to mortality; white players and small players live longer. Our study is useful for devising strategies for health interventions and the proper allocation of resources with respect to the general population.


Introduction
Active participation in sports is generally considered to be beneficial for health and well-being. Protecting the health of athletes is one of the priorities of the International Olympic Committee. The relationship between health and participation in sports has been studied from different perspectives with regard to mortality. Some recent studies have highlighted fatalities among players in

Design, Subjects, and Variables
We performed a retrospective cohort study based on the records of all 3985 active and former NBA players. The cohort included all professional basketball athletes who had played in the NBA from its inception in 1946 to April 2015 (when we began the study). Players who only participated in the ABA league were excluded (299 cases), but players who participated in both NBA and ABA leagues were included. The ABA league was an alternative professional league from 1967 to 1976.
Players were identified from www.basketball-reference.com, which is one of the most comprehensive databases of basketball player information. Players were automatically linked to their height, playing position, place of birth, games, and minutes played in their NBA career, together with the date of their debut and last season played. Playing position was divided into the following categories: Center (C), Center-Forward (C-F), Forward-Center (C-F), Forward (F), Forward-Guard (F-G), Guard-Forward (G-F), and Guard (G). These positions represent a proxy regarding the physical and technical features of players. Regarding place of birth, we differentiated between US and non-US players. The remaining variables of interest were manually recorded player by player. Data on handedness, weight, ethnicity, date of birth, and date of death were added to the database by two of the authors and three independent coders, who received prior training for that purpose. The number of deceased players before April 15, 2015 was 734.

Validation of Variables: Height and Ethnicity
We implemented a sequence of manual data entry using a validation process. In the first stage, researchers entered data on weight, date of birth, and date of death in a sample of 1620 cases.
Two trained coders (Coder 1 and Coder 2) then recorded the remaining cases (1178 and 1187, respectively), after which a third coder (Coder 3) validated the codification. The number of mismatches (disagreements) identified for these three variables was 72, 72, and 49, respectively. We reviewed and resolved all mismatches. Ethnicity was recorded by two independent coders and classified in three categories: African-American, White/Asian, and Mixed. Coders were trained to identify players who were born from parents with different ethnicity, such as Jason Kidd, Tony Parker, Blake Griffin or Aaron Gordon. Once this process was finished, all researchers reviewed the codification and discussed dubious cases. The final categorization was reached by consensus. Coding of handedness was also done manually by two of the authors without any recorded mismatches.
Registration of height required particular attention because differences were found from one source to another. While we initially collected the data from www.basketball-reference.com in April 2015, the height of 43 players had changed when the data were checked again in June 2017. To fix this divergence, we consulted www.NBA.com, where the heights recorded were similar to those obtained in June 2017 from www.basketball-reference.com (39 of 43 coincidences). We decided to use heights from NBA.com for all of these 43 players. The International Basketball Federation (FIBA) also reports the height of participating players; we identified 238 players from the FIBA website who were also in our database and found that the heights reported for FIBA tournaments varied for almost all cases. As there was no way to decide which of the sources was most reliable, we regressed FIBA heights on NBA heights for the 238 players identified and observed a statistically significant constant (7.040) and regression coefficient (0.964) (standard errors were bootstrapped), with an R-square of 0.953. Using these parameters, we estimated the heights of the remaining 3757 NBA players as a proxy of their FIBA height. Consequently, we had two different versions of the height variable.
Finally, having completed the codification and validation process, we investigated the cause of death of those 734 players using the Google search engine and specialist sources such as basketball networks (APBR), digital newspapers (New York Times, LA Times), sports websites (ESPN), and other media resources. Only 443 of the 734 deaths (60.35%) could be clearly assigned to a cause. We codified every cause of death using the International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD-10)-WHO Version for 2016.

Statistical Analysis
For the descriptive analysis, we computed the means of absolute and relative frequencies for categorical variables of interest, and measures of central tendency and statistical dispersion for numeric variables.
To analyze survival times among NBA players, we studied two different response variables. First, survival time was defined as the time elapsed from the end of the NBA career until death. Players who were active on 15 April 2015 were excluded from this analysis, and the survival times of former players alive on 15 April 2015 were treated as right-censored. We estimated survival probabilities and median survival time using the Kaplan-Meier method [25]. We conducted multivariate analysis using the Cox proportional hazards model [26], whose measure of effect size is the hazard ratio, i.e., the ratio of instantaneous risk functions, which is assumed to be constant over time. The potential predictors were the players' physical characteristics and NBA career-related variables. In addition, the model was adjusted for age and year of the players' retirement from the NBA. For the second objective, age at death was modeled as the response variable; this analysis included both active and former players. The survival times of players who were alive on 15 April 2015 were treated as right-censored, and all survival times were treated as left-truncated at the age of the corresponding NBA debut. As before, survival probabilities were estimated using the Kaplan-Meier method and the multivariate analysis was conducted using the Cox proportional hazards model. The potential model predictors were the players' characteristics. In addition, the model was adjusted for the year of the players' first NBA season. Studying age at death permits us to estimate LE as the model-based estimate of the median. For both models, only African-American and white players were considered, and the proportional hazard assumptions of both models were checked using the Schoenfeld residuals [27].
Moreover, to compare the mortality of African-American and white NBA players with that of the general African-American and white US populations, respectively, we computed standardized mortality ratios (SMRs) for the years 2000 through 2014. This measure is the ratio of the observed to the expected number of deaths in the study population, where the yearly expected number of deaths was calculated based on the players' ages on January 1, and the survival probabilities of the same year in the corresponding population.
All analyses were conducted using version 3.4.4 of the R statistical software (Vienna, Austria; http://www.rproject.org/). Specifically, we used the R packages survival [28] and epitools [29] to analyze the survival times and to compute the SMRs, respectively. The R code of the analyses carried out and the data are available as Supplemental Material. This work fulfils the transparency and reproducibility of scientific research.

Ethical Considerations
We did not seek ethical approval for this study, as all information used and reported is freely available via online sources. Table 1 presents characteristics of the players included in this study according to their status (active versus former NBA players), and the proportions all former players who died before 15 April 2015. The mean height and weight of players during this period were 198 cm (SD: 9.32) and 94.8 kg (SD: 12), respectively. A total of 1535 players (38.7%) were White, 2324 (58.7%) were African-American, and 103 (2.6%) were mixed. Regarding player position, 11% were centers, 4.53% center-forwards, 28.7% forwards, 8.3% forward-centers, 4.94% forward-guards,34.3% guards, and 8.11% guard-forwards.

Survival Analysis
We used the follow-up data for the 3504 former NBA players, of whom 687 (19.1%) died before 15 April 2015, to study the time elapsed from the end of their NBA career until death.
The unadjusted nonparametric estimation of the survival function ( Figure 1) shows that the estimated probability of survival to 67 years of age is about 0.15. The estimated median survival time is 54.8 years (95% confidence interval, 95% CI = 53.7-55.9 years). Regarding age of death of NBA players, we analyzed data for all active and former players and computed an unadjusted nonparametric estimation of the survival function ( Figure 2). The estimated median age at death was 81.3 years (95% CI = 80.2-82.5 years). The youngest NBA player to die was Nick Vanos, who was killed at the age of 24 years in an airplane crash. The oldest former NBA player was Ben Goldfadden, who died at the age of 99 years.  Table 2 presents the proportional hazards models for both response variables of interest; these models include all variables that were statistically significantly associated with the survival times of Regarding age of death of NBA players, we analyzed data for all active and former players and computed an unadjusted nonparametric estimation of the survival function ( Figure 2). The estimated median age at death was 81.3 years (95% CI = 80.2-82.5 years). The youngest NBA player to die was Nick Vanos, who was killed at the age of 24 years in an airplane crash. The oldest former NBA player was Ben Goldfadden, who died at the age of 99 years. Regarding age of death of NBA players, we analyzed data for all active and former players and computed an unadjusted nonparametric estimation of the survival function ( Figure 2). The estimated median age at death was 81.3 years (95% CI = 80.2-82.5 years). The youngest NBA player to die was Nick Vanos, who was killed at the age of 24 years in an airplane crash. The oldest former NBA player was Ben Goldfadden, who died at the age of 99 years.  Table 2 presents the proportional hazards models for both response variables of interest; these models include all variables that were statistically significantly associated with the survival times of  Table 2 presents the proportional hazards models for both response variables of interest; these models include all variables that were statistically significantly associated with the survival times of NBA players. According to both models, the height of NBA players is related to survival time. The positive sign of the estimated parameters indicates that instantaneous risk increases with increasing height. In addition, when comparing African-American players to white players, the estimated adjusted hazard ratio is 1.4 for both models, indicating higher instantaneous risk among African-American players. The increase in LE in recent decades is reflected in the negative sign of the parameter estimate for the variables for year of players' last (Model I) and first NBA seasons (Model II). The positive sign of the parameter estimate of the age variable at end of their NBA career (Model I) is expected because it reflects the fact that average survival time is longer the younger the player was when he finishes his career. However, it does not imply that this variable is related to age at death. Further NBA career-related variables, such as number of NBA games, which could have been included in Model I, were not statistically significant. The analysis of the Schoenfeld residuals by means of R's cox.zph function revealed that the proportional hazards assumption holds reasonably well. In the case of Model I, the p-value corresponding to the null hypothesis of proportional hazards was 0.061, and in the case of the Model II, it was 0.548. Model II models the age at death of NBA players and can be used to estimate the median survival time based on the variables included in the model, which can be interpreted as the players' LE. Table 3 shows the estimations (95% CI) for African-American and white NBA players of 2 m tall for different NBA debut years. It can be observed that the estimated LE of NBA players increased steadily throughout the last decades and that in 2010, it was 89 years for white NBA players and 85.8 years for African-American players.  Figure 3 shows the standard mortality ratios computed when comparing the age-adjusted mortality rates of African-American and white NBA players to those of the respective general US populations from 2000 through 2014. For almost all years analyzed, the SMRs are less than 1, implying that NBA players have lower mortality rates than the general population.

Comparison of Yearly Mortality Rates of NBA Players with the General Population
Appl. Sci. 2019, 9, x FOR PEER REVIEW 8 of 13 Figure 3 shows the standard mortality ratios computed when comparing the age-adjusted mortality rates of African-American and white NBA players to those of the respective general US populations from 2000 through 2014. For almost all years analyzed, the SMRs are less than 1, implying that NBA players have lower mortality rates than the general population.

Discussion
We analyzed factors associated with mortality among NBA players and compared the LE of NBA players to that in the general population. To achieve this aim, we analyzed data from the 3985 players who participated in the NBA from its inception in 1946 to April 2015.
After adjusting for age at the end of the season and calendar year, we found that mortality is associated with height and ethnicity. Taller players and African-American players have a higher instantaneous death risk than smaller or white players. In addition, overall instantaneous risk of death has decreased over time because players' LE has increased since the inception of the NBA, regardless of height and ethnicity.
However, for the same height and time period, we observed differences between the LE of white and African-American players (Table 3). African-American players have lower mortality than the general African-American population. There were differences for at least in four of the years between 2000 and 2014, and a trend is also evident (Figure 3).
The mortality risk of African-American players in this cohort was lower than that of the general US African-American male population, and it is not obvious why this healthy worker effect (HWE) [30] is only seen in African-American players. Recall that HWE refers to the tendency of the actively employed to have a more favorable mortality experience than the population at large.
The reason may be related to the progressive equalization of wages between African-American and white NBA players. There have been reports on ethnicity-based salary discrimination in professional basketball, where African-American players were paid less than white players for

Discussion
We analyzed factors associated with mortality among NBA players and compared the LE of NBA players to that in the general population. To achieve this aim, we analyzed data from the 3985 players who participated in the NBA from its inception in 1946 to April 2015.
After adjusting for age at the end of the season and calendar year, we found that mortality is associated with height and ethnicity. Taller players and African-American players have a higher instantaneous death risk than smaller or white players. In addition, overall instantaneous risk of death has decreased over time because players' LE has increased since the inception of the NBA, regardless of height and ethnicity.
However, for the same height and time period, we observed differences between the LE of white and African-American players (Table 3). African-American players have lower mortality than the general African-American population. There were differences for at least in four of the years between 2000 and 2014, and a trend is also evident (Figure 3).
The mortality risk of African-American players in this cohort was lower than that of the general US African-American male population, and it is not obvious why this healthy worker effect (HWE) [30] is only seen in African-American players. Recall that HWE refers to the tendency of the actively employed to have a more favorable mortality experience than the population at large.
The reason may be related to the progressive equalization of wages between African-American and white NBA players. There have been reports on ethnicity-based salary discrimination in professional basketball, where African-American players were paid less than white players for similar levels of performance [31]. However, results from this study indicate reported that the NBA has equitable treatment for players of both ethnicities, at least with regard to veteran free agents, as reported previously [32].
This finding contrasts with the wage gap between African-American and white workers in the general population, which grew from 16.9% in 1979 to 22.5% in 2014 [33]. The relative "ethnicity advantage" of African-American players with respect to the general African-American population is thought to be partly due to salary, owing to the known impact of wage on socioeconomic status (SES). Nonetheless, while SES is the most important factor, it is not the only one that explains racial differences in mortality [34].
The significant effect of ethnicity on mortality could therefore be partly explained by historical salary differences between African-American and white players, at least up to the mid-1980s [31,35]. The effect of wage equalization on players' instantaneous risk of death should be evaluated again in the future when data on the mortality of players who played from the mid-1980s onwards become available); notably, the "racial" salary gap has started to grow again since 2006 [35]. However, it is also possible that other features related to ethnicity could affect death risk. For example, the authors of [36] commented on the results from the study of 566 NBA players carried out by Engel et al. [9]. Left ventricular hypertrophy (LVH) was observed in 27.4% of athletes, with ethnicity identified as a key determinant of the prevalence and geometric pattern of LVH (more prevalent in African-American players). Marked changes in electrocardiographic repolarization have also been demonstrated in athletes of African-American ethnicity [37].
Further, African-American players have a significantly higher likelihood of dying from cardiac disease than white players [38]. Although of great interest, the conclusions derived by Engel et al. contribute to a framework to help to avoid unnecessary exclusion of athletes from competition, rather than to predicting further negative health outcomes [9]. Therefore, factors associated with mortality of basketball players following heart events are still a topic of interest.
While most studies find that taller people have lower rates of CVD than shorter people [39], it is also evident than a notable number of NBA players suffer from heart problems. For example, Larry Bird was diagnosed with atrial fibrillation in 1995, and current players LaMarcus Aldridge, Jeff Green, and Channing Frye have had heart issues [16]. Indeed, increasing height is an independent risk factor for atrial fibrillation, and measured and genetically determined height have been reported to be significantly associated with PR interval and QRS duration [40].
Regarding respiratory diseases, height is a positive predictor of emphysema in patients with chronic obstructive pulmonary disease [41], although conversely, height was found to be inversely associated with mortality [42,43].
Our study has also confirmed that height is positively associated with mortality, as other studies have previously shown [13,38]). Lemez et al. did not find substantive variability in the causes of death between basketball players and the general population [38], highlighting the need for further work. As previously mentioned, height has been found to be positively associated with cancer risk and negatively associated with CVD in other populations [10][11][12]. Better understanding of the special features of basketball players regarding height is needed to analyze the relationship with diseases, most notably CVD, because height is an independent risk factor for atrial fibrillation [40]. While height has been inversely associated with CVD in many studies, numerous basketball players suffer from heart problems and have high risk of sports-related SCD and PE.
However, it has been found that elite and team athletes of aerobic-anaerobic sports live longer than the general population [19,20]. This apparent contradiction between positive and negative factors highlights the need to better understand the causes of mortality in retired basketball players and their LE compared with the general population.
It is also important to mention that the remaining covariables considered when estimating the models were not relevant. Handedness was a special interest because some studies have shown that left-handers have lower longevity than right-handers [44]. However, our results do not support this relationship, as another study has also found [39].
Our Model II makes predictions about the LE of NBA players. For example, the estimated median age at death for former NBA players with characteristics similar to Larry Bird's (height = 205.74 m, white NBA player, and NBA debut in 1979) is 82.8 years (95% CI: 80.1-86).

Limitations and Strengths
The main limitation of our study is the limited knowledge about the life-history of players once they retire. It is well known that some remain active with a healthy lifestyle, while others gain excessive weight. In addition, we did not have access to disease or family data nor the clinical history of the players. Therefore, our results are only suggestive about a general association between mortality and height and ethnicity.
We considered the number of games played as a proxy to measure the volume of exercise, but we do not have tracking data that can inform about the intensity, load or density of exercise. We also had information about the number of minutes played and the number of minutes per game played, but all three variables were highly correlated. Further, we had missing data of minutes played for 344 players, so we only analyzed the number of games played. Very extreme or very intense exercise has been related to mortality [24], but we did not find any association between the number of games played and risk of death. However, many of the players in our sample had careers outside the NBA, playing in other leagues inside and outside United States. While some of these competitions are also very demanding (especially the major European competitions), the level of physical requirement is lower than that for the NBA. However, this absent information was a shortcoming for our analysis.
A considerable percentage of games in the NBA are considered as "close games" [45], where home/away teams are winning by less than 10 points at the end of the 3rd quarter. Specifically, we explored a large data base provided by www.nbastuffer.com; a sample of 2623 regular season and play-off games between 2009 and 2011. The percentage of "close games" was 52.2%. This probably means that the distribution of close games in our whole sample is not homogenous, but these claims should be tested in further research with more comprehensive data. It is known that emotional stressors trigger cardiovascular events [46], but literature has also shown that high drama/intensity experience lived by sports fans can be associated with changes in cardiovascular death rates in the short term, depending on winning (decrease) or losing (increase) [47]. Therefore, it is not clear how highly intense, mentally demanding games in the NBA could influence mortality in the long term, considering the possible moderating factor of winning vs. losing close games. In addition, divergences in intensity between players are not related to the risk of all-cause mortality [48], as has been suggested after comparing National Football League participation as a career player with participation as a replacement player.
Finally, regarding ethnicity, we recognize the complexity of considering this variable in our analysis. As Pedro Gómez pointed out, we have thousands of genes that are invisible except for a few anatomical features such as skin color [49], so categorizing individuals only for one visible feature could be deemed crude. We followed the example of other studies that used the simplistic categorization of White versus African-American using skin color [9]. However, we took precautions to carefully consider and eliminate "mixed ethnicity" from the analysis, in order to take into account players with white and African-American parents (56, 1.6%).
The aim of our study was to analyze overall mortality, but not specific mortality, as Lemez et al. recently did [38]. This requires advanced survival analysis, such as the competing risk method, to analyze the causes of death. However, we believe it would be necessary to reduce the number of missing cases. Lemez et al. [38] had 34.7% of missing cases and in a search, we obtained 39.6% of missing cases. However, it is difficult to know the causes of death of players where information is not available from public sources.
Regarding methodological considerations, we analyzed survival time, rather than a dichotomous response variable (alive or dead). Moreover, we considered the age at death of NBA players to estimate LE. Finally, we used SMRs to compare the mortality of African-American and white NBA players with that in the general African-American and white US population.

Practical Implications
Evaluating mortality among NBA players will be useful for devising strategies for health interventions and the proper allocation of resources with respect to the general population.
In 2015, players' associations began to increase their efforts to offer cardiac screening and to raise awareness about the incidence of these diseases among retired players [50]. The NBA also began to provide information on healthy practices that could help improve their overall health status.

Conclusions
Mortality among NBA players was associated with ethnicity and height. White players and small players live longer. In addition, NBA players have a lower instantaneous risk of death than the general US population. This difference is especially notable among African-American players, with respect to the US African-American population.