A Prospective Study of Health and Risk of Mortality After Spinal Cord Injury

Health Factors and Mortality 

In the aforementioned studies, the health conditions leading to mortality were directly identified from causes of death mostly listed on death certificates. In contrast, there has been limited research that has independently and prospectively measured secondary conditions and other indicators of morbidity and linked them with eventual mortality. We therefore know very little about how the development of a specific health condition may increase the likelihood of mortality or how subtle indicators of change in health may also signal risk of early death. Nevertheless, the success of interventions to enhance longevity is dependent on an understanding of these factors.

Early studies of the association of multiple variables with mortality suggested the psychosocial and vocational outcomes were more highly correlated with mortality than were general health indicators.20, 21 Because the time between collection of the prospective data and determination of mortality status was 11 years and 15 years, respectively, health status could have changed substantially in the years after data collection, as was suggested by a study that used a shorter follow-up interval (4y) but had identified only 22 deceased participants.22 A relatively recent report from the MSCIS in the United States14 found general health ratings were associated with an elevated risk of mortality, but the data on more specific conditions were limited.

Purpose 

The purpose of this study was to identify which health outcomes are associated with elevated risk of mortality after SCI. By focusing on health factors and mortality, this study evaluated the primary component of a hierarchical empirical risk model for mortality23 while controlling for the independent effects of biographical and injury factors.⁎ There are 2 related objectives: (1) to evaluate the extent to which health outcomes, as measured by general indices, indicators of health impact, and secondary conditions, are associated with hazard of mortality after SCI; and (2) to evaluate the extent to which accounting for different sets of predictors enhances our prediction of mortality, moving progressively from injury severity to a comprehensive model that includes all the health factors available in the study.

Our hypotheses were (1) when statistically controlling for biographical and injury characteristics, several types of health factors will be associated with hazard of mortality, including indices of general health, health impact, and secondary conditions; and (2) when building an optimal risk model for mortality, inclusion of health factors will enhance our prediction of hazard for mortality above and beyond that of biographical and injury factors alone.

Participants 

After obtaining approval from the institutional review board, participants were identified from 3 types of records of a specialty hospital in the Southeastern United States: (1) MSCIS patient database, (2) MSCIS registry, and (3) outpatient directory. To be included in the study, participants were adults with traumatic SCI occurring at least 1 year prior to the study that resulted in some residual impairment (those with no neurologic deficits were excluded). Of an original cohort of 1929 patients, 1389 participated (response rate, 72%).

Seventy-four percent were men, and 75% were white. The overall breakdown by sex and race and ethnicity was 54.7% white men, 19.3% white women, 20.2% minority men, and 5.8% minority women. Of the minority participants, 87.7% were African American. Average age at time of injury was 31.8 years (IQR, 20.6–39.8y). At the time of prospective data collection, participants' mean age was 40.7 years (IQR, 30.3–48.9y). They had been injured a mean of 9.8 years (IQR, 3.5–12.2y), and the average education was 13.1 years. The primary etiology was vehicular crashes (50.9%), followed by falls or flying objects (17.3%), acts of violence (12.7%), sports (11.9%), and other (7.2%). Fifty-five percent reported cervical injuries. Just over a fifth (21.4%) reported functional movement below level of injury sufficient to assist ambulation.

Prospective Data Collection Procedures 

Letters announcing the study were sent to all prospective participants, followed by a mailed survey 4 to 5 weeks later. All introductory letters followed U.S. Postal Service standard guidelines to identify changes of address whenever possible. Two additional mail-out attempts and a follow-up phone call were made to nonresponders. Prospective data collection was initiated in July 1997 and ended in April 1998, although nearly all data had been collected by the end of December 1997. To elicit the highest possible response rates, participants were offered a $20 stipend and were made eligible for drawings totaling $1500.

Determination of Mortality Status 

Mortality status was determined an average of 8 years after the primary data collection (December 2005). Persons not found to be deceased by December 31, 2005, were assumed to be alive. To ensure a comprehensive determination of mortality, we elected to use both the NDI of the National Center for Health Statistics24 and the SSDI of the Social Security Administration.25 Although neither data source is reported to be 100% inclusive,26, 27, 28 researchers often use the NDI and the SSDI to determine mortality. NDI provides access to information on decedents through a centralized computer index by searching death records provided by state offices. NDI provides information on the name of the state where death occurred, the death certificate number, and the date of death for each subject searched and identified. NDI death records are available approximately 16 months after the conclusion of a given year (ie, NDI records were not available for 2004–2005). Among persons in the MSCIS whose full name and Social Security number were known, sensitivity and specificity of using the SSDI were 92.4% and 99.5%, respectively.5 In a later study of 371 known decedents, the SSDI correctly identified almost 87% of decedents age 50 years and older, 69.8% of decedents age 30 to 49 years, and only 34.6% of decedents younger than age 30 years. Factors that influenced sensitivity were older age, male sex, white race, married status, and survival greater than 1 year after injury.29

Measures 

Several instruments were used to measure the health factors and secondary conditions, including subsets of items from the BRFSS30 and LSQ-R.31 The OAHMQ32 was used to measure depressive symptoms. Selected items were also developed for the study.

Several types of health outcomes were measured that fall into the following broad categories: (1) general health indicators, (2) treatment variables, and (3) secondary conditions. The secondary condition variables fell into 3 general classes: symptoms of SCI complications (particularly symptoms of infections, eg, urinary tract complications, sweats, chills), irreversible conditions (eg, amputations), and reoccurring conditions (eg, pressure ulcers, injuries, depression). To minimize the potential for an inflated type I error rate, the biologic plausibility of each health variable was considered prior to analysis. Whereas the first 2 classes of variables, general health indicators and treatment variables, are indicators of overall health, secondary condition variables are more viable targets for intervention.

Table 1 summarizes the variables to be used in the study. The first variable (self-health rating) may be best conceptualized as the summation of health, not linked to any specific secondary condition, event, or health variable. It was measured by a single BRFSS item: self-rating current health. Two other health impact variables were measured by 2 additional BRFSS items: days in poor physical health and days in poor mental health, each of which required participants to indicate the number of days out of the past 30 for which their physical or mental health was not good. The CDC has used these items extensively and has existing data available for normative purposes. Three treatment variables were used from the LSQ-R. Each of the 3 LSQ-R items requires participants to indicate over the previous 12 months the number of nonroutine physician office visits, (2) the number of hospitalizations, and (3) the number of days hospitalized.31 Several types of secondary conditions were assessed. Symptoms of secondary conditions were measured by items reflecting complications of SCI. These included fevers, sweats and chills, and UTIs (self-report based on diagnoses by physicians). Participants were asked to indicate the number of occurrences of each of these in the past year using ordinal rankings (0, 1–2, 3–6, 7–12, ≥13). A summary measure (summated score) was created. Irreversible secondary conditions reflected historical events that could not be reversed (eg, upper-limb or lower-limb amputation), events that led to a major nonrecurrent complication (upper-extremity or lower-extremity fracture, DVT), or the accumulation of conditions that, although in theory reversible, in practice rarely occur (eg, upper-limb or lower-limb contractures, curvature of the spine). There were 8 items, each of which asked participants whether the condition had ever occurred. The items included (1) blood clots (DVT), (2) amputation of a lower extremity (leg, foot), (3) amputation of an upper extremity (hand, arm), (4) curvature of the spine (more difficulty sitting up straight), (5) broken bone in a lower extremity (leg, foot), (6) broken bone in an upper extremity (hand, arm), (7) contracture in hip, knee, or ankle, and (8) contracture in elbow or shoulder.⁎

Table 1. Variables Used in the Data Analysis, Type of Variable, and the Instrument From Which They Were Selected

	Biologic Domain	Type of Variable⁎	Instrument
	Variable
	A. General health indicators
	General health rating	Ordinal: scale A	BRFSS
	No. days poor health over past 30 days	Continuous	BRFSS
	No. days poor mental health over past 30 days	Continuous	BRFSS
	B. Treatments
	Nonroutine physician visits over past year	Ordinal: scale B	LSQ-R
	Hospitalizations (any reason) over past year	Ordinal: scale B	LSQ-R
	Days spent hospitalized over past year	Ordinal: scale C	LSQ-R
	C. Secondary conditions
	C1. Symptoms of SCI complications (in past year)
	Stomach pain and distention (bloating)	Ordinal: scale D	SCI health survey
	Bowel accidents	Ordinal: scale D	SCI health survey
	Rectal bleeding	Ordinal: scale D	SCI health survey
	Urine leaking or accidents	Ordinal: scale D	SCI health survey
	Fevers	Ordinal: scale D	SCI health survey
	Sweats and chills	Ordinal: scale D	SCI health survey
	UTIs (must be diagnosed by a doctor)	Ordinal: scale D	SCI health survey
	No. of UTI-like symptoms	Continuous	Derived as sum of fevers, sweats/chills, and diagnosed UTIs
	C2. Irreversible conditions
	Report fractures to lower or upper extremities since injury	Binary	SCI health survey
	Report amputations to lower or upper extremities since injury	Binary	SCI health survey
	Fracture or amputation since injury	Binary	Derived indicator of any fracture or amputation since injury
	Reported blood clots, curvature of the spine, or contracture since injury	Binary	Derived from items on SCI health survey
	C3. Recurrent conditions
	Pressure ulcers
	Since injury, how many pressure ulcers were severe enough to require medical attention	Ordinal: scale B	SCI health survey
	Pressure ulcer that required attention	Binary	Derived as having at least 1 pressure ulcer
	Since injury, how many surgeries to repair pressure ulcers	Ordinal: scale B	SCI health survey
	Since injury, what best describes pressure ulcer history	Categorical: scale E	SCI health survey
	History of frequent pressure ulcers	Binary	Derived indicator representing people who always seem to have ulcers, often requiring surgery or hospitalization
	Subsequent injuries
	In past year, no. of injuries that required medical care	Ordinal: scale B	SCI health survey
	Since injury, no. of injuries that required medical care	Ordinal: scale B	SCI health survey
	Since injury, no. of times hospitalized because of injuries	Ordinal: scale B	SCI health survey
	Depressive symptoms
	Probable major depression	Binary: scale F	OAHMQ

Scale A: poor (1), fair (2), good (3), very good (4), excellent (5). Scale B: 0 (0), 1 (1), … , 9 (9), 10+ (10). Scale C: 0 (0), 1 (1), … , 29 (29), 30+ (30). Scale D: How many times did you have the following problems? 0, 1−2 (1.5), 3−6 (4.5), 7−12 (9.5), 13+ (14). Scale E: A, Have never had any pressure ulcers; B, had pressure ulcers immediately after injury, but rarely, if at all, since initial rehabilitation hospitalization; C, get pressure ulcers about every couple years; D, get at least 1 pressure ulcer a year; and E, always seem to have ulcers, often requiring surgery or hospitalization. Scale F: OAHMI scores: normal depressive symptomatology (OAHMI scores <6); clinically significant symptomatology (OAHMI score range, 6 to <11); and probable major depression (OAHMI score range, 11−22). Binary depression variable of OAHMI scores ≥11.

⁎ For each of the following scales, the value provided in parentheses is the ordinal score used for the analysis. If no score is given (and the type of variable is categorical), the corresponding indicator (ie, dummy variable) was constructed.

In contrast, recurrent secondary conditions may have repeated cycles. We assessed 3 types of conditions: pressure ulcers, subsequent injuries, and depressive symptoms. Pressure ulcers were defined as “open sores in pressure areas, such as your tailbone, ischium, heel, elbows. They are usually caused by pressure but may also be caused by friction or shearing (rubbing), moisture, burns, or falls.”33(p1259) Participants were to indicate (1) the number of pressure ulcers they had in the past year, (2) the number of days they were forced to reduce their sitting time as the result of a pressure ulcer over the past year (they were presented 8 multiple-choice categories), (3) whether they had a pressure ulcer at time of the study, (4) the number of surgeries since the onset of their SCI to heal pressure ulcers, and (5) which of 5 categories best described their pressure ulcer history (categories were described along a continuum from nonrecurrent to recurrent).

Developed in collaboration with the National Center for Injury Prevention of the CDC, subsequent injuries were defined in the following manner: “The following questions relate to INJURIES, including broken bones, burns, or lacerations (cuts). Injuries happen as the result of some type of mishap or event, such as a fall, collision, motor vehicle wreck, or act of violence.”34(p1504) Participants were asked the number of times in the past year they had been injured seriously enough to receive medical care in a clinic, emergency department, or hospital and whether any of these injuries had been serious enough to require treatment in a hospital for at least 1 night. Subsequent injuries were reported in 19% of participants over a 12-month period, with 27% of participants who reported at least 1 injury also reporting 1 or more injury-related hospitalizations.34

The 22-item OAHMQ32 was used to measure depressive symptoms. The OAHMQ was developed to measure depression in older adults and among people with physical disabilities by including few items that reflect physical or vegetative symptomatology, because these types of items often invalidate commonly used measures of depression and other clinical syndromes. Scores 11 or higher were considered to indicate probable major depression. Kemp and Adams32 reported test-retest correlations were found to be .87 (P<.001) and α values were .93 (P<.001). In terms of validity, sensitivity and specificity were reported to be .93 and .87, respectively. The OAHMQ was also compared to 2 other depression scales with established validity and reliability, the Geriatric Depression Scale and the Symptom Checklist−90−Revised. Correlation for the OAHMQ and both measures was .70 (P<.001). The OAHMQ has been used in previous research with SCI.35

Data Analyses 

A 3-stage hierarchical strategy to model building was employed to identify the association of each health variable with mortality and to define an optimal set of health predictors of mortality. Cox proportional hazards modeling was used with the number of days between the survey and event (ie, mortality) as the dependent variable. The censoring date was December 31, 2005, the date up to which mortality status could be verified.

During the first stage of analysis, a base model consisting of biographic and injury characteristics, including sex, race (white-minority), age, years lived since injury, and injury severity, was specified (table 1 summarizes the variables and their coding). Injury severity was defined by a combination of 2 items that were used to generate 5 categories that were similar, but not equivalent, to those frequently reported in the SCI mortality literature.⁎ The groups were defined as a combination of injury level and neurologic completeness of injury, with injury level broken down into 3 categories (C1-4, C5-8, T1-S5) among those who did not report motor functional recovery sufficient for ambulation. Among those who had recovery sufficient for ambulation, we further broke this group into those with cervical injuries and those with noncervical injuries. Table 2 summarizes the breakdown of participants according to this categorization.

Table 2. Breakdown as a Function of Injury Level and Neurologic Completeness and Ability to Ambulate

	Category	No.	Percent⁎
	C1-4, nonfunctional	176	13.2
	C5-8, nonfunctional	407	30.5
	Noncervical, nonfunctional	469	35.2
	Cervical, ambulatory	142	10.7
	Noncervical, ambulatory	139	10.4
	Unclassified (not used in analysis)	56

⁎ Percentages exclude the 56 observations not classified.

The second stage of the analysis focused on adding single, biologically plausible health outcome variables to the base model, thereby screening each of these potential predictors for inclusion in the final stage model. All variables significant at the α equal to .10 level of significance were considered for subsequent modeling.36 Once variables had been screened within a biologic domain (eg, recurrent secondary conditions), those identified were then assessed for multicolinearity within a biologic domain, and backward selection was used to identify the key predictor variables for each biologic domain. The variables identified by using this process are summarized in table 3.

Table 3. Cox Proportional Hazards Analysis for the Preliminary and Full Models

	Biologic Domain	Stage 2 Results			Selected for Stage 3	Stage 3 Results
	Variable	HR⁎	95% CI	P		HR†	95% CI	P
	Core adjustment variables‡
	Functional status
	C1-4, nonfunctional	NA	NA	NA	Yes	4.25	1.88−9.64	<.01
	C5-8, nonfunctional	NA	NA	NA	Yes	2.87	1.30−6.34	<.01
	Noncervical, nonfunctional	NA	NA	NA	Yes	2.58	1.17−5.68	.02
	Cervical, ambulatory	NA	NA	NA	Yes	1.29	0.50−3.33	.60
	Noncervical, ambulatory (referent)	NA	NA	NA	Yes	1.00	ND	ND
	Caucasian vs non-Caucasian	NA	NA	NA	Yes	1.14	0.82−1.60	.43
	Sex	NA	NA	NA	Yes	1.22	0.87−1.73	.25
	Age at injury	NA	NA	NA	Yes	1.06	1.05−1.07	<.01
	Years since injury	NA	NA	NA	Yes	1.04	1.02−1.06	<.01
	A. General health indicators
	General health rating	0.62	0.53−0.72	<.01	No§	ND	ND	ND
	Poor health days	1.02	1.01−1.04	<.01	No§	ND	ND	ND
	Poor mental health days	1.03	1.01−1.04	<.01	No§	ND	ND	ND
	B. Treatments
	Nonroutine doctor visits	1.08	1.04−1.12	<.01	Yes	NS	ND	ND
	Hospitalizations	1.16	1.09−1.23	<.01	No	ND	ND	ND
	Days spent in hospital	1.03	1.02−1.05	<.01	Yes	1.02	1.00−1.03	.03
	C1. Symptoms for SCI complications
	Stomach pain/distention	1.02	1.00−1.05	.05	No	ND	ND	ND
	Bowel accidents	0.99	0.96−1.03	.71	No	ND	ND	ND
	Rectal bleeding	0.99	0.96−1.03	.59	No	ND	ND	ND
	Urine leaking or accidents	0.99	0.97−1.02	.61	No	ND	ND	ND
	Fevers	1.05	1.02−1.08	<.01	No	ND	ND	ND
	Sweats and chills	1.05	1.02−1.08	<.01	No	ND	ND	ND
	UTIs	1.07	1.03−1.10	<.01	No	ND	ND	ND
	No. of infection symptoms	1.03	1.02−1.04	<.01	Yes	1.02	1.00−1.03	.03
	C2. Irreversible conditions
	Fractures to extremities	1.50	1.05−2.15	.03	No	ND	ND	ND
	Amputations to extremities	3.69	2.14−6.37	<.01	No	ND	ND	ND
	Fracture or amputation since injury	1.81	1.31−2.52	<.01	Yes	1.78	1.27−2.51	<.01
	Other irreversible secondary conditions	1.24	0.94−1.64	.13	No	ND	ND	ND
	Pressure ulcers
	No. of ulcers needing medical care since injury	1.15	1.09−1.21	<.01	No	ND	ND	ND
	At least 1 ulcer needing care	1.92	1.43−2.62	<.01	Yes	NS	ND	ND
	Surgeries to repair ulcers	1.20	1.13−1.27	<.01	Yes	1.16	1.08−1.23	<.01
	History of frequent ulcers	3.68	2.18−6.19	<.01	No	ND	ND	ND
	Subsequent injuries
	Injuries in the past year needing medical care	1.13	1.00−1.26	.04	No	ND	ND	ND
	Injuries since injury needing medical care	1.18	1.10−1.26	<.01	Yes	NS	ND	ND
	Hospitalized for injury since injury	1.16	1.09−1.25	<.01	No	ND	ND	ND
	Depression
	Probable major depression	2.21	1.66−2.94	<.01	Yes	1.86	1.36−2.53	<.01

Abbreviations: CI, confidence interval; NA, not applicable; ND, no data to enter into cell; NS, variable was eliminated from the stage 3 model using backward selection.

⁎ HRs from a Cox model predicting the time, measured in days, from survey to either mortality or censoring (12/31/05) and adjusted for core (biographic and injury) variables listed in the first biologic domain. † HRs for the stage 3 final model are adjusted for all other variables in which an estimate is provided. Dashes indicate that the variable was not considered for stage 3 analysis. ‡ HRs for core adjustment variables not shown for stage 2 results because there is variability of the estimates depending on which of the candidate variables is used in the analysis. § General health indicators, while significant predictors in stage 2, were not considered for stage 3 of the analysis. They were used, however, for sensitivity analyses and alternative model specifications.

The final stage of the analysis formulated a Cox proportional hazards model that consisted of the base model in addition to the variables identified in stage 2 of the analysis. After assessment for multicolinearity, backward elimination was used to identify the final fitted model. The general health indicators domain was not considered for the backward elimination; however, the general health rating and physical health rating from the BRFSS were included in alternative (exploratory) models. We chose to focus on secondary conditions because they have more direct implications for prevention.

The proportional hazards assumption was assessed using the Schoenfeld residuals37 and found to be tenable. The fit of the model was assessed using the likelihood ratio test and the c statistic.38 The likelihood ratio test was used to calculate the Nagelkerke pseudo R2.39 The value of the c statistic is closely related to the area under a receiver operating characteristic curve and is interpretable as the probability that the cases (ie, deaths) have higher risks as measured by the linear component of the regression model. Accordingly, a value of 0.5 is for chance prediction, and the discrimination of the model is improved as the c value approaches 1.0.40, 41 Once the final model was determined, all pairwise interaction terms of the secondary conditions were included in a new model to assess goodness of fit further. A Wald linear contrast indicated these interaction terms were not needed in the model (P>.50), and accordingly, these interaction terms were removed. All model building was conducted using the SAS.a The validation of the proportional hazards assumption and the estimation of the c statistic were performed using Stata.b

Study Implications 

Diminished life expectancy after SCI is one of the most well established findings in SCI research. Yet how often does increased risk of mortality enter into routine outpatient assessments, patient conferences, or physician or health provider discussions with people with SCI? The current study does not address this question, yet it certainly indicates that this is a question that needs to be addressed. How can we intervene to reduce the risk of mortality if it is not even a topic discussed with our patients with SCI?

We have sufficient research to direct interventions toward early identification of risk on the basis of the presence of multiple health indicators. Pressure ulcers, UTIs, amputations or fractures, and the development of depressive disorders are all predictive of early mortality, and the presence of these conditions should serve as a red flag for intervention. Some interventions can and should be incorporated into routine assessments, the minimum of which is to inform patients with SCI regularly of the association between their health status and early mortality so they have heightened awareness of risk.

Although we have not demonstrated a causative relationship (only an association) between these factors and mortality, secondary conditions with the strongest associations with early mortality are logical targets for intervention. One type of screening that appears to be particularly important is that of depressive symptoms and the possibility of a major depressive disorder. Because most people who find their way to outpatient clinics present with problems, health care providers and outpatient clinics may assume that depressive symptoms are a natural comorbidity of health complications. A minimum of brief screening is required, sometimes followed up by a full assessment by a licensed psychologist, to rule out more dangerous depressive diagnoses. Because the final model suggests that a broad constellation of risk factors, reflective of diverse underlying constructs, are predictive of mortality, a more holistic approach to promote better health may be needed to reduce mortality.

Study Limitations 

This study had several limitations that limit the generalizability of the findings. First, we have established an association of health outcomes only with mortality, not causality. Second, the data were heavily left censored. Participants entered into the analysis an average of just under 10 years after SCI onset, so a substantial number of potential participants may have died prior to initiation of the data collection. Similar to the most recent MSCIS analysis,22 there were no data in the first postinjury year when many factors such as ventilator dependency, injury severity, and age exert their greatest influence on the likelihood of mortality. Third, NDI data were not available from January 1 to December 31, 2005, so there may be a small number of deaths that were not detected because of the 92.4% sensitivity of the SSDI among deceased persons.14 Fourth, we did not collect data on ventilator status, which is highly correlated with probability of mortality, particularly during the first few years after SCI onset.22 However, we do not anticipate a significant number of these participants given the left censoring (ie, because they have the shortest longevity, and many would have died before enrollment).

There were additional factors that limit the extent to which our design had the power to evaluate fully the relationships between health factors in mortality. First, the sample size was smaller in scope than that used with studies from the model SCI systems in the United States (these studies have focused primarily on biographic and injury characteristics). Although this limitation is balanced with the more detailed data on risk and protective health factors, larger sample sizes would have greater power to identify significant predictors of mortality. Second, recall bias may have influenced the accuracy of certain reports of health conditions during the prospective data collection. Inaccuracy in self-reported health would be an error in measurement and would weaken our ability to identify relationships between health and mortality. Last, risk and protective factors were only measured once with as much as an 8-year lag between collection and determination to mortality status. If data on these variables were collected more recently, they would serve as more precise predictors of mortality. For instance, probable major depression was a prominent predictor of mortality, yet these symptoms may represent chronic dysphoria consistent with a major depressive disorder (highly stable) or transient symptoms more related to situational circumstances. Similarly, the absence of these at one point in time does not preclude their later development (ie, a portion of participants reporting no symptoms would have developed symptoms at a later point in time). Therefore, it may be best to consider the current estimates as conservative.

Future Research 

Three types of research are suggested by the current findings. First, given that this study was based on a multistage model of prediction of mortality, and health and secondary conditions are only 1 step of factors from the model, future research should investigate additional risk and protective factors including behaviors, psychologic characteristics, and environmental factors. Investigation of these factors is particularly important because identification of risk prior to the development of complications will enhance prevention strategies. Second, we need to identify risk and protective predictors for the health factors that related to early mortality. This essentially speaks to the same issue: identification of risk and protective factors early in the chain of events leading to early mortality. The third recommendation for future research is both the most important and most challenging: development and evaluation of interventions to increase longevity after SCI. Our findings clearly may serve as at least a partial foundation for the development of interventions in that they identify people at high risk for early mortality and the health conditions that lead to the elevated risk of mortality. It is only through continued epidemiologic and clinical intervention research that long-term survival will likely be enhanced.