Timing and Predictors of Mild and Severe Heat Illness among New Military Enlistees

Abstract

Purpose

Heat illnesses are important and potentially fatal conditions among physically active individuals. We determined predictors of heat illness among enlistees in a large military population experiencing common physical activity patterns.

Methods

We estimated the adjusted odds of mild and severe heat illness associated with demographic, health-related, and geographic factors among active-duty, United States Army soldiers enlisting between January 2011 – December 2014 (N=238,168) using discrete-time multivariable logistic regression analyses.

Results

We observed 2,612 incident cases of mild heat illness (MHI) and 732 incident cases of severe heat illness (SHI) during 427,922 person-years of follow-up, with a mean and median of 21.6 and 21 months per subject. During the first six duty months, 71.3% of the MHIs and 60.2% of the SHIs occurred, peaking at month two. The odds of MHI quadrupled among those with prior SHI (OR=4.02, 95% confidence interval [CI]: 2.67 – 6.03). Body mass index (BMI) extremes increased risk substantially (ORs at BMI ≥30: for MHI, 1.41, CI 1.19 – 1.67; for SHI, 1.94, CI 1.47 – 2.56; ORs at BMI <18.5: for MHI, 1.50, CI 1.01 – 2.21; for SHI, 2.26, CI 1.16 – 4.39). Tobacco use was associated with a 55% increase (CI: 1.37 – 1.77) in MHI odds. The odds of MHI increased if taking NSAIDs, opioids or methylphenidate stimulants. Lower age and lower entry aptitude scores were associated with progressively increased MHI odds.

Conclusion

The majority of heat illnesses occurred at the outset of service, indicating the need for focused prevention methods at the initiation of military duty. Prior heat illness, BMI extremes, medications and tobacco use represent potentially actionable risk factors to address by education, policy, and/or clinician intervention.

INTRODUCTION

Heat illnesses comprise situation- and person-dependent physiologic responses that may occur when an individual is under thermal stress. In terms of severity and pathophysiology, heat illness diagnoses constitute a range of clinical syndromes, from relatively milder illnesses such as heat cramps to much more serious classic and exertional heat stroke (1,2). Rates of heat illness in the United States are mainly documented in the emergency setting and vary widely across the nation. The 2014 state-level rate of heat-related emergency department cases per 100,000 persons in the continental US ranged from 4.6 to 32.6 (3). However, reported rates of heat illness may be underestimated (4,5).

Heat illness risk factors include being elderly or suffering chronic illness, which can increase the risk of classic heat stroke (6,7), or being an active youth, placing one at risk for exertional heat stroke (7, 8). Having a high body mass index (BMI) (9,10) or taking certain medications may also increase heat illness risk. Relevant medications include non-steroidal anti-inflammatories (NSAID; 11), of which one formulation was recently shown to increase heat stroke risk in an animal model (12), as well as stimulants and alpha blockers (13). Prior heat illness is also believed to be a risk factor for subsequent, potentially more-severe events (14).

Military personnel are a good population for studying incident heat illnesses. Exertional heat illnesses, in particular, represent a substantial occupational hazard during rigorous military service (15). Before entering service, potential soldiers are screened for health problems including serious prior heat illness, which is a disqualifying condition (16). After service entry, new soldiers progressively engage in similar types and levels of strenuous activity in known locations with diverse and sometimes extreme climates, which leads to elevated risk of exertional events.

Observations beginning at the outset of military service could therefore enable the capture of truly incident heat illness events, among which exertional etiologies are presumed to predominate, and under relatively standardized conditions. Further, large data reserves are maintained on soldiers, to include information on multiple heat illness risk factors. Improved knowledge about the timing and predictors of heat illness in this population could translate to enhanced prevention efforts, reduced morbidity and lives saved.

Therefore, the aims of this study were to retrospectively observe a population of new military enlistees in longitudinal data in order to: (1) determine the temporal patterns of incident heat illness events during initial exposure to the strenuous activity associated with the military life; (2) assess the association of various demographic, health-related, and environmental exposures with mild (MHI) and severe (SHI) heat injuries; and, in particular, (3) examine the contribution of geographic location to heat illness risk.

METHODS

Subjects

We conducted a retrospective cohort study by using the Stanford Military Data Repository. The analysis leveraged its capture of digitally recorded health encounters, dispensed prescription medications, demographics and physical locations of all active-duty U.S. Army enlisted soldiers (i.e., persons not initially commissioned as officers) serving between 2011 and 2014. We restricted the study population to individuals who began duty and served for any duration during this time span (N=238,168). Subjects were observed in a person-month-based panel data structure for incident heat illness by using a time-to-event approach. Observation began at service entry and continued until censoring due to an outcome event, military discharge, or the end of the available data in December 2014, whichever occurred first.

Data sources were employed to capture a sufficient range of outpatient and inpatient care to identify both the heat illness events and the potential predictors of interest. To produce the longitudinal analytic file for this study, medical record extracts were merged at the person-month level with comprehensive administrative and personnel information from official sources; identifiers were removed prior to research activity. These sources, which are maintained by the Department of Defense, are described in Table 1. This research was approved by the Institutional Review Board at the Stanford University School of Medicine. The project received secondary review and approval by the Human Research Protections Office of the Defense Health Agency. Informed consent was deemed not applicable to this population-based, effectively de-identified dataset on subjects from across the total US Army.

Table 1.

Descriptions of the datasets used.

Defense Manpower Data Center (DMDC; 17)

Master File – Official Department of Defense (DoD) demographic and military service data and Armed Forces Qualification Test (18) scores Transaction File – Official DoD records of military duty status changes, such as discharges from service

Military Health System Data Repository (MDR; 19)

Combined Ambulatory Professional Encounter Record: Official DoD archives of diagnoses made during all digitally-captured outpatient care in military facilities Tricare Encounter Data, Non-Institutional: DoD archive of all diagnoses made during outpatient care in civilian facilities that was reimbursed by military health coverage Standard Inpatient Data Record: DoD archive of all diagnoses made during digitally-captured inpatient care in military facilities Tricare Encounter Data, Institutional: DoD archive of all diagnoses made during inpatient care in civilian facilities that was reimbursed by military health coverage Clinical Data Repository Vitals File: Did archive of height and weight readings at CAPER (outpatient) encounters in military facilities Pharmacy Detail Transaction Service: DoD archive of details of medications dispensed by military or civilian pharmacies if under military health coverage

Medical Operational Data System (MODS)

Periodic Health Assessment: Self-report and clinician-entered data from required annual health screenings (20) at military health facilities eProfile: Data from the official electronic archive of formal duty and activity restrictions or "profiles" (21), with associated clinical reasons

Digital Training Management System (DTMS)

Army Body Composition Program: Readings from required, biannual height and weight assessments (22) Army Physical Fitness Test (23) data: Scores from required, biannual assessments of upper and lower body strength and endurance and cardiorespiratory endurance

Data Organization

The dependent and independent variables used in the regression models are listed below with brief descriptions.

Dependent Variables

The primary outcomes of interest were mild (MHI) and severe (SHI) heat illness, which were established as discrete endpoints with dedicated analyses. Of the relevant syndromes, we noted that heat stroke has generally received the most attention in the research literature. Most heat stroke cases in this active, generally young study population were presumed to represent exertional rather than classic heat stroke. However, to identify heat illness events, we chiefly employed International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes that identify problems including heat stroke, but do not distinguish between classic and exertional forms. Although we did not have access to detailed chart content on the history of each case to assist us in discriminating such cases, a review of our clinical data, together with our temporary and permanent duty restriction (eProfile) data indicated that, due to clinician and/or diagnosis coder choices in the population studied, heat illness diagnoses other than heat stroke were sometimes associated with serious sequelae. Associated outcomes included hospitalization and enduring occupational limitations assigned by clinicians.

We therefore classified MHI and SHI events by severity after examining the total health and administrative trajectories available in the SMDR data, as described below. For MHI and SHI, an individual with either outcome was provided a "1" value for a binary variable in the person-month of the incident event observed in out- or inpatient care. Cases constituted incident SHI when the subject was first diagnosed with heat stroke per ICD-9-CM code "992.0," or when diagnosed with another heat illness type documented with a "992" diagnosis code prefix (such as heat exhaustion or heat syncope), if there was evidence of severe effects related to the event. These effects included occupational limitations exceeding 60 days in length that were specifically assigned due to the incident heat illness occurrence, and/or associated clinical sequelae. The selected sequelae were a hospital admission, renal disorders, blood dyscrasias, electrolyte disorders, and/or rhabdomyolysis. See supplementary Appendix 1 for additional information on the derivation and usage of these data [see Document, Supplemental Digital Content 1, Usage and derivation of mild (MHI) and severe heat illness (SHI) outcome events].

Incident cases of MHI were dichotomously defined by the first observed non-heat stroke heat illness diagnosis (i.e., any diagnosis with a "992" prefix other than "992.0," which represents heat stroke) that occurred without associated lengthy work limitations, hospital admission, or any of the clinical comorbidities described above in our SHI definition. See the online Appendix 1 for additional information [see Document, Supplemental Digital Content 1, Usage and derivation of mild (MHI) and severe heat illness (SHI) outcome events].

Independent Variables

Demographic factors

Data from official personnel records provided control variables to address gender, age, race, and marital status for each subject.

Army Physical Fitness Test (APFT) scores

Run times have been found to predict heat illness among military recruits in past studies (24,25). The US Army administers biannual physical fitness tests to its soldiers that include timed push-ups, sit-ups and a two-mile run (23). We used the total APFT score, which includes scores assigned for run times, as a general indicator of strength and fitness including cardiovascular capacity. The approach was also taken because APFT scores rather than raw performance data are pre-weighted for gender and age (23). APFT data were arranged using a categorical variable indicating the score from the last fitness test in each subject's longitudinal data.

Entry aptitude test scores

A higher cognitive ability has been associated with better general health outcomes in adolescents (26). We hypothesized that heat illness risk in this population of mostly young recruits may be inversely related to individual cognitive capacity reflected by aptitude testing scores, perhaps reflecting an increased capacity to understand and therefore engage in behavioral thermoregulation. To address cognitive capacity as a potential heat illness risk factor, we utilized data from the Armed Forces Qualifying Test (AFQT) portion of the Armed Services Vocational Aptitude Battery (18). The test is administered to all prospective, enlisted US military service applicants. The AFQT consists of language and mathematics tests and constitutes the main indicator of qualification for service and particular occupations. We categorized the AFQT score based on the frequency distribution of scores in the study population. AFQT scores were absent for those who transitioned from enlisted status to officers after entry and for an additional 14 enlisted subjects, which necessitated a dedicated category to cover these subjects.

Months of military service and early service

We included service time as a continuous variable accounting for months of duty to address the impact of exposure time, and support statistical objectives to be discussed. Because of the event distributions to be reported, a further dichotomous covariate was included to address whether the subject was or was not in the first four months of service.

Medications

The availability of prescribed selected medications was identified using official pharmacy data (Table 1) that included quantities and usage directions for medications identified by American Hospital Formulary System (27) therapeutic classes. Potential use was based solely on the medical provider's intent to treat the patient as reflected in prescription instructions; actual patient compliance was unknown to the study. Each binary medication variable was assigned a "1" value in a month in which the subject was previously prescribed medication enabling at least 30 days of potential use, such that the potential usage period concluded either in the month with the "1" value, or at the end of the prior month.

Medication variables with a "1" value therefore always identified agents dispensed in a month previous to the month of a potentially associated outcome. This approach was taken to preserve causal inference in this person-month-based dataset, wherein actual patient use of medication was unobservable. Otherwise, it would be more likely that an agent dispensed in the same month as an outcome could have first been taken after the outcome. The number of MHI outcomes among medication recipients permitted the assessment of associations with recently dispensed NSAIDs, opioids, alpha blockers, amphetamines and methylphenidates. Insufficient outcomes among those who received opioids, alpha blockers, and methylphenidates for these variables were available to be included in the SHI model.

Body Mass Index (BMI)

Height and weight readings drawn from health encounter and physical examination data, as well as from required biannual readings (22), were used to compute BMI. BMI was a running value constituting the last reading taken as of each observed person-month. We categorized these values by using the standard definitions of underweight, normal, overweight and obese BMI ranges (28), which are quantified in Tables 2 and 3. We also provided a distinct category for those without BMI data.

Table 2.

Raw study population numbers (percentages)of newly enlisted US Army soldiers (N = 238,168) within selected subgroups with results of chi square tests comparing factor distributions for those who were and were not diagnosed with mild (MHI) and severe heat injury (SHI) during 2011–14.

Distributions among total study population (N = 238,168)	Diagnosed with MHI?		Diagnosed with SHI?
	No 235,556 (98.9)	Yes 2612 (1.1)	No 237,436 (99.7)	Yes 732 (0.3)
Gender	P < 0.001		P < 0.001
Male	198,992 (99.1)	1856 (0.9)	200,309 (99.7)	539 (0.3)
Female	36,564 (98.0)	921 (2.0)	37,127 (99.5)	193 (0.5)
Race	P = 0.030		P < 0.001
White	167,542 (98.9)	1841 (1.1)	168,937 (99.7)	432 (0.3)
Black	52,697 (98.8)	627 (1.2)	53,067 (99.5)	256 (0.5)
Asian/Pacific Islander	11,879 (99.0)	119 (1.0)	11,959 (99.7)	39 (0.3)
Other or unknown	3438 (99.3)	25 (0.7)	3473 (99.9)	5 (0.1)
Age, years	P < 0.001		P < 0.001
≤20	59,035 (97.8)	1306 (2.2)	59,485 (99.5)	306 (0.5)
21 – 22	69,055 (99.2)	595 (0.9)	69,671 (99.7)	174 (0.3)
23 – 24	46,396 (99.3)	310 (0.7)	46,786 (99.8)	102 (0.2)
≥25	61,070 (99.4)	401 (0.7)	61,494 (99.8)	150 (0.2)
Mean / median age	23.1 / 22	21.5 / 20.5	23.1 / 22	22.2 / 21
Marital status	P < 0.001		P < 0.001
Married	71,157 (99.4)	433 (0.6)	71,806 (99.8)	128 (0.2)
Never married	161,110 (98.7)	2143 (1.3)	162,281 (99.6)	600 (0.4
Formerly married	3289 (98.9)	36 (1.1)	3349 (99.9	4 (0.1)

Table 3.

Odds ratios (ORs), statistical significance indicators^A and 95% confidence intervals (CIs) for independent variables^B included in discrete-time logistic regression models employed to assess the adjusted odds of mild (MHI) and severe heat injury (SHI) among newly enlisted United States Army soldiers (N = 238,138) during 2011–14. ORs for non-displayed reference categories for binary covariates were 1.00.

Factor	MHI		SHI
	OR	CI	OR	CI
Female gender (referent: male)	2.14***	1.95 – 2.34	1.66***	1.40 – 1.98
Race	Referent: white		Referent: white
White	1.00	-	1.00	-
Black	0.94	0.86 – 1.04	1.72***	1.46 – 2.03
Asian or Pacific Islander	0.93	0.77 – 1.12	1.24	0.89 – 1.73
Other or not declared	0.92	0.62 – 1.37	0.79	0.33 – 1.90
Age, years (referent: ≥25)
≤20	1.23**	1.08 – 1.40	0.89	0.71 – 1.12
21 – 22	1.16*	1.02 – 1.33	0.93	0.73 – 1.17
23 – 24	1.05	0.90 – 1.22	0.92	0.71 – 1.18
Marital status (referent: married)
Never married	1.11	0.99 – 1.25	1.29*	1.05 – 1.59
Formerly married	1.52*	1.08 – 2.14	0.58	0.22 – 1.58
Pay grade	Referent: lower grade		Referent: higher grade
Lower enlisted	1.00	-	2.76	0.61 – 12.4
Higher pay grades	1.44	0.76 – 2.70	1.00	-
Armed Forces Qualifying Test (18) score, 0 – 100 (referent: ≥75)
≤42	1.26***	1.12 – 1.42	1.04	0.84 – 1.30
43 – 56	1.14*	1.01 – 1.27	0.92	0.74 – 1.14
57 – 74	1.02	0.90 – 1.14	0.93	0.75 – 1.15
No data available	1.24	0.54 – 2.86	6.62*	1.36 – 32.3
Army Physical Fitness Test (23) score, 0 – 300	Referent: ≥270		Referent: 245 – 269
<215	1.10	0.87 – 1.40	1.34	0.84 – 2.12
215 – 244	0.99	0.79 – 1.25	1.17	0.75 – 1.81
245 – 269	0.84	0.66 – 1.08	1.00	-
≥270	1.00	-	1.32	0.83 – 2.10
No data available	1.21	0.99 – 1.48	2.20***	1.52 – 3.17
Prior heat injury of other type (referent: none)	4.02***	2.67 – 6.03	1.77	1.00 – 3.13
Body Mass Index (BMI), kg/m2 (referent: 18.5 – 24.99)
<18.5	1.50*	1.01 – 2.21	2.26*	1.16 – 4.39
25 – 29.99	1.10*	1.01 – 1.19	1.29**	1.10 – 1.51
≥30	1.41***	1.19 – 1.67	1.94***	1.47 – 2.56
Unknown	0.23***	0.16 – 0.33	0.30***	0.16 – 0.56
Tobacco use (referent: none)	1.55***	1.37 – 1.77	1.19	0.95 – 1.49
NSAID use (referent: none)	1.31*	1.05 – 1.64	0.84	0.49 – 1.44
Opioid use (referent: none)	1.92*	1.08 – 3.41	N/A	N/A
Alpha blocker use (referent: none)	6.09	0.84 – 44.1	N/A	N/A
Amphetamine use (referent: none)	0.70	0.18 – 2.86	2.93	0.73 – 11.8
Methylphenidate use (referent: none)	5.68*	1.41 – 22.9	N/A	N/A
First four months of service (referent: later service)	2.97***	2.60 – 3.39	1.79***	1.41 – 2.26
Season (referent: winter)
Spring	5.90***	4.56 – 7.63	5.55***	3.60 – 8.55
Summer	22.1***	17.3 – 28.2	16.3***	10.8 – 24.6
Fall	4.95***	3.82 – 6.42	4.83***	3.12 – 7.45
Location groups designated primarily by latitudeC (referent: Group 1, highest latitude)
Group 2	1.87***	1.46 – 2.39	1.24	0.78 – 1.96
Group 3	4.74***	3.72 – 6.04	8.25***	5.42 – 12.6
Group 4	2.78***	2.19 – 3.54	2.28***	1.49 – 3.49
Hawaii	2.59***	1.76 – 3.81	4.23***	2.34 – 7.63
Months of service (continuous value)	0.97***	0.96 – 0.98	0.99	0.98 – 1.00

^A. Statistical significance: *≤0.05; **<0.01; ***<0.001

^B. The models included all variables listed, and additionally controlled for the presence of temporary and permanent duty restrictions from the eProfile system (18).

^C. See Appendix 2 for additional details on the location groups.

Tobacco use

Given its potential impact on a wide range of health outcomes, tobacco use was identified from self-report at health care encounters and annual health screenings. We employed a running, binary indicator to identify any prior statement of tobacco use as of each person-month.

Prior heat illness

In the analysis of each endpoint, MHI and SHI, the prior presence of the other event as each was previously defined was included as a dichotomous independent variable in any person-month before which the other outcome was observed. For example, hypothetically consider a subject who experienced two heat illnesses in the study's observed time: a MHI event as we defined it, and six months prior to that event, a SHI. In the analysis of incident MHI events, the SHI event would constitute a "1" value for prior heat illness in each person-month after the SHI occurrence. In the SHI analysis, the variable for prior heat illness would carry a "0" value in the months preceding the SHI, since the SHI was the first observed heat illness. This approach was taken to preserve and account for realistic health trajectories in which any possible sequence of heat illness events might occur, and allowed us to quantify the contribution of prior heat illnesses to the adjusted odds of later incident occurrences of outcomes of the other severity level. See the supplementary Appendix 1 online for additional information on usage and derivation of data addressing heat illness events [see Document, Supplemental Digital Content 1, Usage and derivation of mild (MHI) and severe heat illness (SHI) outcome events].

Location of service

Soldiers spend their initial 10 weeks of service at a Basic Combat Training (BCT) site. BCT is immediately followed by job-specific Advanced Individual Training (AIT) at either the same or a different physical site (29), and subsequent service at initial duty assignments worldwide. We therefore included a covariate to estimate the impact of the prevailing climate during each observed person-month as subjects moved through different locations in their early careers. Each subject's monthly location was classified using official service records and evidence of physical locations determined by facilities at which health care was received.

Sites were organized into groups by using a primarily latitude-based approach (see Table, Supplemental Digital Content 2, North latitudes and selected additional information for United States Army locations grouped for regression analyses). The categorical covariate accounted for US and non-US sites across the globe, including Europe and Korea. Figure 1 graphically displays the US sites, and includes notation of BCT and/or AIT locations. Because of its unique location, climate and relatively large Army population (30), Hawaii was provided a dedicated category. Recognizing the potential importance of factors such as altitude and proximity to bodies of water, in addition to latitude, the locations in each group, especially those at the geographic margins of potential groups, were extensively tested for consistency in the regression analysis to arrive at the final location combinations.

Figure 1.

United States Military Service Locations.

Season of the year

A categorical covariate designated the season in which each person-month was observed. The seasons were defined as winter (December – February), spring (March – May), summer (June – August) and fall (September – November).

Duty restrictions

Opioids and NSAIDs may be prescribed to treat injuries that require medical duty restrictions. The presence of such conditions and/or the associated restrictions could affect activity levels and otherwise relate to the risk of heat illness. The models included covariates for temporary and permanent activity restrictions to control for this potential effect.

Statistical Analysis

Statistical power computations confirmed power exceeding 80% for associations of interest. Overall rates of MHI and SHI and rates per observed month of service were computed. Initial descriptive analyses included cross-tabulations chi square tests for differences in the frequency distributions of the chosen predictors among those with and without outcomes.

After preliminary analyses, we computed the adjusted odds ratios for MHI and SHI associated with the selected covariates using dedicated, multivariable discrete-time logistic regression models for each outcome. Variable category multicollinearity and outcome rarity distribution in some variable categories dictated reconfiguration of the relevant covariates for the SHI model.

In combination with the censoring plan and person-period data structure, the use of the service time covariate in logistic regression facilitated a similar outcome risk assessment as that produced by Cox proportional hazards models (31). Our methods using logistic regression provided odds ratios that are mathematically equivalent to hazard ratios in terms of providing a conditional heat illness probability value associated with observations as of a given time point.

All analyses were conducted using Stata 14 software (StataCorp, College Station, Texas).

RESULTS

The 238,168 subjects were observed for a total 427,922 person-years of service. Mean observed time per subject was 21.6 months (median: 21; standard deviation: 13.7; range: 1 – 48). We observed a total of 2,612 incident MHI and 732 incident SHI events, which affected 1.1% and 0.3% of the study population, respectively. Of the MHI events observed, 1,862 (71.3%) occurred during the first six duty months, as did 441 (60.2%) of the SHI events.

Figure 2 displays the timing of MHI and SHI events during the observed service time of up to 48 months in terms of events per 100,000 subjects. Rates of each event peaked in month two of service and tapered substantially thereafter. Simple descriptive analyses indicated that new soldiers who experienced heat illness differed in terms of multiple demographic and health behavior-related factors compared to those who had not experienced heat illness (Table 2).

Figure 2.

Incident Mild and Severe Heat Illness Events per 100,000 Newly Enlisted US Army Soldiers in each Service Month.

Table 3 displays the results of the discrete-time, multivariable logistic regression models employed to estimate adjusted associations between the selected factors and the incident appearance of each heat illness type. Several demographic factors were associated with heat illness events. The adjusted odds of either heat illness type were higher among women compared to men. The sole statistically significant effect associated with race was a 72% increase in the adjusted odds of SHI associated with black race when compared with Caucasians (95% confidence interval [CI]: 1.46 – 2.03, P < 0.001). The odds of MHI decreased with increasing age, but no differences in the odds of SHI among differing age categories were noted. Subjects who had never been married were at 29% higher odds of SHI (CI: 1.05 – 1.59, P = 0.016) and formerly married subjects were at 52% higher odds of MHI (CI: 1.08 – 2.14, P = 0.017) than married subjects.

No statistically significant findings were noted for military pay grades. The variable was retained in the model due to the substantial effect sizes nonetheless revealed. Among the other military-specific factors, we noted several findings. Lower AFQT scores conferred progressively-higher MHI odds. Those with missing AFQT data were at much higher adjusted odds of SHI (odds ratio [OR]: 6.62; CI: 1.36 – 32.2; P = 0.020). No statistically significant findings were noted for the physical fitness score ranges we explored, but those with missing APFT data were at higher odds of SHI (OR: 2.20; CI: 1.52 – 3.17; P < 0.001).

Among clinically-focused variables, a prior SHI was associated with a four-fold increase in the odds of MHI (CI: 2.67 – 6.03; P < 0.001). A prior MHI conferred a 77% increase in the odds of later SHI, with marginal significance at the 95% confidence level (CI: 1.00 – 3.13; P = 0.052). The highest and lowest categories of BMI demonstrated relatively strong adjusted associations with each heat illness type. BMI ≥ 30 kg/m² was associated with a near-doubling of the odds of SHI (OR: 1.94; CI: 1.47 – 2.56; P < 0.001), and BMI < 18.5 was associated with an even slightly higher odds ratio (OR: 2.26; CI: 1.16 – 4.39; P = 0.016), when either was compared to those with normal BMI. Effect size estimates for these BMI strata were each lower, though still significant, in the MHI analysis (OR for BMI ≥ 30: 1.41; CI: 1.19 – 1.67; P < 0.001; OR for BMI < 18.5: 1.50; CI: 1.01 – 2.21; P = 0.042). The absence of BMI data was associated with substantially reduced odds of either heat injury severity type.

Tobacco use was associated with a 55% increase in the adjusted odds of an MHI (CI: 1.37 – 1.77; P < 0.001), but the association was not statistically significant for SHI. The strongest medication effect with statistical significance was seen for methylphenidate stimulants, associated with over five times the odds of MHI among its users (OR: 5.68; CI: 1.41 – 22.9; P = 0.015). NSAID use was not associated with SHI, although these agents were associated with an increase in the odds of MHI (OR: 1.31; CI: 1.05 – 1.64). Opioids nearly doubled the adjusted odds of MHI (CI: 1.08 – 3.41; P = 0.027). The use of alpha blocker agents conferred a six-fold increase in the odds of MHI, though with marginal significance and a wide confidence interval that arose from low usage rates in this population of new soldiers (CI: 0.84 – 44.1; P = 0.074). The presence of duty restrictions (not reported in Table 3) was not associated with statistically significant changes in heat illness odds.

Consistent with the unadjusted information in Figure 2, the adjusted odds of both heat illness types increased during the first four months of duty, in which soldiers typically engaged in basic and advanced training. Duty in the summer was associated with the greatest odds increase for either illness type, but spring and fall were also associated with higher odds when compared to winter. Interestingly, observation in the predominately warm Group 3 location belt demonstrated substantially higher effects estimate than did either the more southern Group 4 or Hawaii.

DISCUSSION

Several critical findings in this work have implications for military and athletic training policy. The first such finding is the very early peak incidence of both heat illness types in the first few months of military service, indicating that efforts to reduce risk should be optimized at the earliest possible time in an Army soldier's career. The second finding is that prior heat illness predicts later events. This observation suggests that while US Army prevention policies warn of potential risk among those with prior heat illnesses (14), these individuals might be afforded even greater attention. In particular, return-to-activity policies after heat illness should be assessed for proper implementation, and potentially modified if heat illness rates continue unchanged in a given setting. A third finding with policy implications is that rates of both MHI and SHI were lower in Hawaii and consistently warm or hot areas than in some higher-latitude locations with cooler climates. This variability may provide an opportunity to review training or other policies that could systematically differ across locations, perhaps as a product of prevailing climates.

Another possible explanation for the risk differences observed across location groups is increased acclimatization among subjects in the lowest-latitude areas and Hawaii. The issue of acclimatization to heat and activity deserves focused attention because it is an established approach leveraged to reduce heat injury risk among athletes (32) and soldiers (14,33). Exposing soldiers to graduated strenuous activity may be the most actionable approach at any geographic location because it may increase heat tolerance more than passive exposure to heat itself (34). Also, service members may originate from areas and/or during seasons where actual heat acclimatization may be challenging or impossible. Therefore, engaging in progressively increasing exercise intensities over a period of time before or immediately upon beginning military service may be helpful in reducing the early-service heat illness rate, particularly among soldiers who originate from cooler climates.

A related finding was that although we did not find statistically significant effects for quantified levels of physical fitness, missing APFT data were associated with increased odds of SHI. APFT data are more likely to be missing among soldiers with medical limitations. As a result, this finding could represent a deconditioning effect, in that soldiers might suffer SHIs at increased rates upon return to activity after medical problems, further emphasizing the importance of acclimatization for the resumption of physically demanding duties. As discussed in the description of independent variables, we employed total APFT scores versus the run times examined in prior research (24,25). It is worthwhile to note that, in addition to our focus on the general fitness indicator, sensitivity analyses were conducted to test whether run times demonstrated a different relationship with heat illness when compared to total APFT score. Results were unchanged.

Recent US military health surveillance reported higher unadjusted rates of heat stroke among men than women, but reported higher unadjusted rates of other heat illnesses among women than men (35). We found that females had higher adjusted odds for both MHI and SHI, which we did not define by simply dividing events into heat stroke versus other diagnoses. Our results may also differ from other analyses because ours were derived from regression analyses that controlled for a range of demographic and clinical factors, and our study only included Army soldiers.

Our finding is generally consistent with other research suggesting that females are less tolerant to heat during exercise than men (36). An additional explanation for gender differences in heat illness risk could be lower average aerobic capacity among women, as lower aerobic capacity has been reported to be a risk factor for heat illness (37). In our study, however, we did not observe any statistically significant effects for Army Physical Fitness Test scores, thus this possibility may merit further study.

Our findings of increased MHI risk at lower ages in this young study population suggest the potential for multiple causal pathways. A tendency for younger soldiers to engage in more frequent and/or strenuous activity may explain increased risk among younger soldiers when controlling for service time and military rank. However, the entire population studied is generally expected to engage in relatively similar types and levels of physical training and other forms of regular occupational exertion. The age effect therefore deserves more study to determine the mechanisms for the association seen in this population. Mechanisms might include improved behavioral thermoregulation with age, and decreased resilience among young adults today that is revealed when exposed to exertion and heat. We additionally observed differing adjusted risk profiles for MHI and SHI associated with marital status, a finding of uncertain etiology and utility.

Prior heat illness, observed extremes of BMI, tobacco use, opioids, NSAIDs and stimulants were also found to be risk factors for heat illness. These findings constitute further actionable evidence for those responsible for the safety and health of active individuals, and may assist targeted education on heat illness prevention. In particular, being underweight may represent an overlooked factor when assessing heat illness risk. Otherwise, the very low odds of either heat illness type in the absence of BMI data suggest a resilience effect, in that BMI data were mostly captured at health encounters. Soldiers who rarely need health care may be the least likely to suffer from problems including heat illness.

Because of the potential association of opioids and NSAIDs with duty restrictions stemming from injuries, the findings of increased risk for those on these medications may in fact be due to deconditioning followed by a rapid return to activity. To guard against this phenomenon, clinicians may consider writing duty restrictions to specifically include a phased return to activity in such cases. However, the opioid finding deserves further research that we are pursuing.

Our findings on medication effects also suggest that policies on the use of certain agents may need to be adjusted for active populations. NSAIDs may deserve particular attention in light of risk factor concerns in prior research (11) and emerging evidence from an animal model indicating a risk increase associated with one such agent, indomethacin (13). Other actions by the military might include broader, more standardized applications of the military's practice of using markers on uniforms to identify the level of recent hydration and any history of prior heat illness (39), and reconsidering assignment locations and selection or retention of individuals with prior heat illness or who chronic use of high-risk agents.

A prior study examining tobacco use and heat injury in an active military population did not find a statistically significant association between these factors (38), but utilized a relatively small study population with more limited tobacco use. In our data, we noted that although heat illness rates were lower among those with tobacco use in a cross-sectional analysis, the converse association was observed when using longitudinal data in the multivariable analysis. Given our prior finding of an association between tobacco use and exertional rhabdomyolysis among active-duty soldiers (40), we suggest that there should be greater attention to tobacco cessation efforts in active populations. Additional efforts could focus on the availability of tobacco products or even consideration of the eligibility of tobacco users for service, recognizing practical difficulties that the most stringent measures may face.

The finding for increased MHI odds associated with low aptitude scores supported our theory that lower cognitive ability may increase the probability of heat illness. This association may exist due to the cognitive skills required to consistently engage in the individual behaviors involved in heat illness prevention, including hydration, activity modulation and early symptom recognition. Coaches and leaders should consider the audience when instructing their charges on heat illness precautions, as persons with differing cognitive capacities may process and use information very differently. The high effect size for absent aptitude data deserves further study, as the main reason for its absence was the transition to officer status.

This study had a number of strengths and limitations. Strengths included the incorporation of relatively standard data on a very large population with a substantial degree of clinical and geographic detail. However, the somewhat low number of SHI events presented some challenges to variable construction and regression modeling. Otherwise, the military represents an excellent population to study because its members experience a relatively similar type and level of exertion and activity, often in warm climates. We do acknowledge that the unique culture and environment of the military population studied may limit the ability to generalize findings to certain other specific populations. We would suggest that this population's composition of young adults engaging in relatively consistent strenuous activity, while often wearing protective and other equipment such as helmets, may be comparable to college and professional athletes and those in certain other occupations, such as firefighters.

An additional potential limitation is the ICD-9 coding that was largely used to define outcomes; diagnosis coding can be subject to clinician bias and adjustment by administrative staff. A related concern is the potential for differing associations to be present for exertional and non-exertional heat illness. We suspect the vast majority of cases in this very active population were exertional in nature due to the rigor of military training. However, ICD-9 diagnosis codes used to define outcomes do not specify whether or not an event is exertional in nature. Given the nature of the available data and the very large study population, chart reviews or other means needed to differentiate exertional and non-exertional cases were not possible.

We further recognize that once engaged in occupation-specific activities during regular duty at their post-initial training assignments, exposures among soldiers of differing military occupations may have been substantial with respect to exertion and heat exposure. The potential impact of exposure time for persons in a given occupation might therefore differ from those in another in ways that would require focused study. Yet, given the common exposures that soldiers experience during BCT, and the fact that the highest heat illness rate by far occurred during that time (see Figure 2), we suspect minimal bias was introduced into our analysis by this phenomenon. Finally, we acknowledge the inability to confirm medication compliance, which may deviate substantially from clinical directions.

In conclusion, this longitudinal, retrospective cohort study identified multiple, potential intervention options for clinicians, leaders or policy makers seeking to mitigate heat illness risk in a range of venues. The findings provide strong evidence that modifiable and actionable factors do exist and provide potential opportunities to reduce heat illness rates. The majority of heat illnesses occurred at the outset of service, indicating the need for focused prevention methods in the earliest phases of military duty/training. Prior heat illness, BMI extremes, medication and tobacco use represent potentially actionable risk factors to address by education, policy, and/or clinician intervention.

Supplementary Material

Appendix 1

Appendix 1: Usage and derivation of mild (MHI) and severe heat illness (SHI) outcome events.

Appendix 2

Appendix 2: North latitudes and selected additional information for United States Army locations grouped for regression analyses.