Trends, Cost, and Mortality from Sepsis after Trauma in the United States: An Evaluation of the National Inpatient Sample of Hospitalizations, 2012–2016

Abstract

Objective:

Identification and outcomes in patients with sepsis have improved over the years, but little data are available in patients with trauma who develop sepsis. We aimed to examine the cost and epidemiology of sepsis in patients hospitalized after trauma.

Design:

Retrospective cohort study

Patients:

National Inpatient Sample (NIS)

Interventions:

Sepsis was identified between 2012–2016 using implicit and explicit International Classification of Diseases, Ninth and Tenth Revision codes. Analyses were stratified by injury severity score (ISS) ≥15. Annual trends were modeled using generalized linear models. Survey-adjusted logistic regression was used to compare the odds for in-hospital mortality, and the average marginal effects were calculated to compare the cost of hospitalization with and without sepsis.

Measurements and Main Results:

There were 320,450 (se = 3,642) traumatic injury discharges from U.S. hospitals with sepsis between 2012 and 2016, representing 6.0% (95% CI: 5.9%−6.0%) of the total trauma population (N = 5,329,714; se = 47,447). In-hospital mortality associated with sepsis after trauma did not change over the study period (p>0.40). In adjusted analysis, severe (ISS≥15) and non-severe injured septic patients had an odds ratio of 1.39 (95% CI: 1.31, 1.47) and 4.32 (95% CI: 4.06, 4.59) for in-hospital mortality, respectively. The adjusted marginal cost for sepsis compared to non-sepsis was $16,646 (95% CI: $16,294, $16,997), and it was greater than the marginal cost for severe injury compared to non-severe injury $8,851 (95% CI: $8,366, $8,796).

Conclusions:

While national trends for sepsis mortality have improved over the years, our analysis of NIS did not support this trend in the trauma population. The odds risk for death after sepsis and the cost of care remained high regardless of severity of injury. More rigor is needed in tracking sepsis after trauma and evaluating the effectiveness of hospital mandates and policies to improve sepsis care in patients after trauma.

INTRODUCTION

Sepsis is the leading cause of in-hospital mortality and is the most expensive inpatient condition treated in the United States (US).^1–3 Sepsis diagnoses continue to increase, in part due to better recognition and coding by providers.^{4, 5} Observational studies in sepsis treatment show benefits in early treatment, and national mandates require process measures for early detection and treatment to further curtail the high mortality and cost of care.^{2, 6} Sepsis, however, remains a heterogonous syndrome with differing outcomes among patient cohorts and subtypes.^{7, 8} Little data exist on the cohort of patients with trauma who develop sepsis, including how sepsis identification and outcomes change over time. Much of the data are limited to single center or regional data.⁹

Patients with trauma are a distinct cohort suggesting incidence and outcomes from sepsis may differ from non-trauma hospitalizations.¹⁰ In general, patients with trauma are relatively younger and have fewer comorbidities than other hospitalized patients. The implications of prevention and early recognition are different because sepsis is typically acquired during hospitalization, after the onset of trauma. The epidemiology of sepsis and sepsis-related mortality in trauma are sparse and include a German trauma registry study prior to 2011 and few other studies from nationally representative data.^{11, 12}

We aimed to provide nationally representative data of all trauma hospitalizations from the US National Inpatient Sample (NIS) using different coding methods for sepsis to better understand the epidemiology of sepsis after trauma. Costs, hospital outcomes, and trends were examined across a five-year study period. We hypothesized the following: (1) cost of care in sepsis was higher than cost of care for severe trauma; and (2) survey-adjusted proportions for sepsis and mortality case-rates were variable based on case-identification methods.

METHODS

Data Source and Design

Primary analyses were performed using the United States Agency for Healthcare Research and Quality (AHRQ) Healthcare Cost and Utilization Project’s (HCUP) National Inpatient Sample (NIS). HCUP NIS is the largest all-payer (Medicare, Medicaid, Private, and Uninsured) inpatient care database in the US that is designed to be representative of all community hospitals for national estimates on the population of inpatients. Community hospitals are defined as short-term, non-federal, and non-rehabilitation hospitals. The NIS is drawn from a sampling frame that contains hospitals comprising more than 95% of all discharges from statewide data organizations in a complex survey design. The makeup of NIS is a weighted sample of the State Inpatient Databases in a single-cluster design stratified on geographic area, urban/rural, ownership, teaching status, and bed size.¹³ The NIS is also standardized across years to facilitate trend analyses.

The American College of Surgeons Trauma Quality Improvement Program (TQIP) is a second data source targeted as a trauma quality database for national comparisons; however, several limitations exist that preclude inclusion for this study. ¹⁴ First, sepsis is identified only if an obvious source of infection with bacteremia and two or more of the following are present: (1) temperature greater than 38°C or less than 36°C; white blood cell count greater than 12,000/mmᵌ or greater than 20% immature bands, hypotension, evidence of hypoperfusion, anion gap or lactic acidosis or oliguria, or altered mental status. The sepsis definition limits the sensitivity and likely excludes many patients with sepsis, especially blood culture-negative cases. We found a crude rate of sepsis at less than 0.5% across the study period in TQIP (Supplemental 1a). Second, TQIP is a convenience sample and not nationally representative data and without cost-to-charge data or individual ICD codes. Additional limitations of TQIP in comparison to NIS are detailed in Supplemental 1b.

Patient Characteristics and Setting

In 2012, NIS was redesigned to optimize national estimates with a sample of discharge data from all HCUP-participating hospitals rather than all discharge data from a sample of hospitals. The population of hospitalizations with trauma was identified by a principal or first International Classification of Diseases, Ninth and Tenth Revision, Clinical Modification (ICD-9/10) for acute injury as previously described.^15–17 ICD codes for late effects of injuries, effects of foreign body, burns, certain early complications of trauma, poisoning by drugs, toxins and other effects were excluded.^16–17 The final study population was comprised of adult hospitalizations (≥ 18 years) with a primary trauma between 2012 and 2016.

Sepsis Definition and Infection Types

Sepsis was identified using explicit and implicit ICD codes put forth by Angus et al.^18, If explicit codes for sepsis, severe sepsis or septic shock were found, then the hospitalization was classified as Angus-positive. If not, all ICD-9/10 diagnosis codes were reviewed for an infection code. If an infection code was present, then the ICD-9/10 diagnoses and procedure codes were examined for codes for acute organ dysfunction. If both an infection and an acute organ dysfunction code were found, then the patient was classified Angus-positive for sepsis. Otherwise, the hospitalization was Angus-negative for sepsis.¹⁹ In subgroup analysis, only explicit codes (sepsis, severe sepsis, and septic shock) were examined. Our study included previously published explicit sepsis codes that used general equivalence mappings to crosswalk ICD-9 into ICD-10 codes and incorporated codes used for the SEP-1 quality metric.²⁰ Codes for post-operative sepsis were included as well.

The Center for Disease Control (CDC) and International Sepsis Forum Consensus Conference definitions were followed for classifying the source of infection in sepsis cases.²¹ Causative organisms were organized using ICD codes on the species level with pathogens grouped into eleven distinct categories (Gram-positive cocci, Gram-positive rods, Gram-negative cocci, Gram-negative rods, anaerobes, atypical bacteria, yeasts and fungi, viruses, other, unknown, or no pathogen).^22,23 Further, infections were categorized by organ system into respiratory (lower and upper), cardiovascular, central nervous system, urinary tract, skin and soft tissue, gastrointestinal tract, reproductive system, bones and joints, eye, ear, oral, and other. ^22,23

Injury Severity Score

The injury severity score (ISS) was assigned using ICD-9/10 diagnosis code and an ISS scoring program made publicly available by HCUP (http://ideas.repec.org/c/boc/bocode/s457028.html). Using this program, the ISS ranges from 1 (least severe) to 75 (most severe). The HCUP ISS data element was based on an anatomical scoring system that provides an overall score for patients with multiple sites of injury. Each injury was assigned an Abbreviated Injury Scale (AIS) score, allocated to one of six body regions [Head, Face, Chest, Abdomen, Extremities (including pelvis), and External]. Serious injury for any given body region was an AIS >2. Only the highest AIS score in each body region was used. The three most severely injured body regions had their scores squared and added together to produce the ISS score.

We examined the association for increased risk of mortality across ISS cutpoints. We modeled the potential non-linear association between rate of in-hospital mortality and ISS using restricted cubic splines with five equally spaced knots.²⁴ The odds ratio and corresponding 95% confidence interval for each ISS level were calculated by applying the delta method.²⁵ This approach has been implemented in the context of fractional polynomials.²⁶ We showed mortality risk began to increase at an ISS of 15 (Figure 1). Our statistical derivation for ISS ≥15 where the odds ratio for in-hospital mortality began to increase is also supported by prior studies showing its association with outcomes after trauma.^25–28 Therefore, we chose ISS ≥15 to define severe trauma for our analyses.

Figure 1.

Distribution of Odds Ratios of In-Hospital Mortality by Injury Severity Scores (ISS) (n=5,329,714)

Analyses

The primary outcomes were all-cause mortality and cost of hospitalization between sepsis and non-sepsis trauma patients. Results used survey-adjusted counts, proportions, means, with associated standard errors (se), and 95% confidence intervals (95% CI). Age- and sex-adjusted year-to-year variability for sepsis and mortality proportions between 2012 and 2016 were modeled with testing for the assumption of linearity using an approach for multiyear survey data.²⁹ We included the interaction between stratum and year to account for geographic region.

To examine risk factors associated with sepsis, a survey-adjusted generalized linear model (GLM) was used and the following candidate variables were examined: demographics, comorbidities with the Elixhauser comorbidity score³⁰, hospital characteristics, and injury characteristics. A Least Absolute Shrinkage and Selector Operator (LASSO) method was applied for variable selection, and model fit across different domains of data were examined by the Akaike information criterion. Multicollinearity of variables were measured by the Variance Inflation Factor and the interaction between sepsis and injury severity was also assessed.

Hospital submitted charges to HCUP were available for the period 2012–2016 to perform cost analysis. Hospital reported charges were converted to costs using the HCUP Cost-to-Charge Ratio files. Each file contains hospital-specific cost-to-charge ratios or a weighted average of hospitals with similar characteristics and from the same State. The Hospital Care component of the Personal Health Price Index published by CMS was used to inflate hospital charges incurred in earlier years to 2016 US dollars.^31,32

The GLM with gamma family and log link function were used to model the cost variable. The average marginal effects were calculated to estimate and compare the cost of hospitalization with and without sepsis adjusted for all confounders. Medians and interquartile ranges and non-parametric tests were used to describe and compare cost variables. Analyses were conducted using Stata with survey features (version 15) and R “survey” package (Version 3.3.1, http://ww.r-project.org). The study was approved by the Loyola University Chicago Institutional Review Board as exempt.

RESULTS

Trends for case-rates of sepsis and in-hospital mortality after trauma (2012 – 2016)

There were 320,450 (se = 3,642) traumatic injury discharges from U.S. hospitals with implicit and explicit sepsis codes between 2012 and 2016, representing 6.0% (95% CI: 5.9%−6.0%) of the total trauma population (N = 5,329,714; se = 47,447) during the study period. The survey-adjusted proportion of sepsis using only explicit ICD codes was 1.2% (95% CI: 1.1%−1.3%). There was no linear trend in age- and sex-adjusted proportions for sepsis in trauma hospitalizations when using both implicit and explicit codes (p=0.67); however, there was a small but increasing trend for explicit sepsis (p<0.01) (Figure 2). For all-cause mortality, no linear trend was noted across both definitions for sepsis (p>0.4) (Figure 3). The trends in all-cause mortality for trauma sepsis were in contrast to a decreasing linear trend in the US population of non-trauma hospitalizations with sepsis (N= 150,839,913, p<0.01) (Supplemental Figure 1).

Figure 2.

Trauma Hospitalizations with Sepsis between 2012 and 2016

Total hospitalizations for each year for specified cohort (trauma vs. non-trauma)

Figure 3.

Mortality in Trauma Hospitalization with Sepsis Between 2012 and 2016

Survey-adjusted proportions with interaction between year and stratum of region.

Baseline characteristics and marginal cost of care in trauma hospitalizations with and without sepsis

Trauma hospitalizations with sepsis compared to trauma hospitalizations without sepsis were older (66.3 vs. 60.1; p<.01) with a larger proportion on Medicare insurance (59.3% vs. 46.5%; p<0.01), a higher ISS (13 vs. 8; p<.01), and a greater proportion with injuries for each body region (p<.01 for all comparisons except extremities). The sites of infection with the greatest frequencies in the sepsis group were respiratory (39.6%) and genitourinary (31.4%), and these were also the sites that had the greatest frequencies with organ dysfunction (Table 1). Trauma patients with sepsis had a longer length of stay (16.0 vs. 5.0 days; p<.01) and a greater proportion with in-hospital mortality (9.5% vs. 2.2%; p<.01) compared to non-septic trauma patients. The median cost of care in trauma hospitalizations with sepsis was nearly three-fold greater than traumas without sepsis (Table 1).

Table 1.

US National Estimates of Trauma Hospitalizations with and without sepsis (N=5,329,714)

Characteristics	Sepsis		No Sepsis		p-value
Hospitalizations, n (%)	307,620	5.8%	5,024,615	94%	<.01
Age, mean (SE)	66.3	0.2	60.1	0.1	<.01
Female, n (%)	147,025	47.8%	2,452,925	48.8%	<.01
Elixhauser, mean (SE)	3	0.03	2	0.01
Race/Ethnicity, n (%)					<.01
White	219,745	71.4%	3,493,560	69.5%
Black	29,525	9.6%	502,090	10.0%
Hispanic	25,065	8.1%	483,730	9.6%
Other	15,745	5.1%	260,945	5.2%
Missing	14,010	4.6%	224,205	4.5%
Insurance, n (%)					<.01
Medicare	182,425	59.3%	2,335,830	46.5%
Private	36,730	11.9%	542,460	10.8%
Medicaid	58,775	19.1%	1,327,420	26.4%
Self-pay	14,285	4.6%	417,335	8.3%
No charge	1,240	0.4%	34,230	0.7%
Other	13,545	4.4%	355,560	7.1%
Missing	525	0.2%	9335	0.2%
AIS >2, n (%)					<.01
Head	86,865	28.2%	709,060	14.1%
Face	740	0.2%	5,730	0.1%
Thorax	49,700	16.2%	414,025	8.2%
Abdomen	1,885	0.6%	8,440	0.2%
Extremities	32,635	10.6%	406,125	8.1%
External	320	0.1%	3,020	0.1%
ISS, mean (SE)	13	0.11	8	0.04
Site of Infection, n (%)					<.01
Gastrointestinal	26,470	8.6%	25,475	0.5%
Respiratory	121,735	39.6%	109,530	2.2%
Other	78,845	25.6%	135,545	2.7%
Cardiovascular	86,995	28.3%	218,830	4.4%
Genitourinary	96,600	31.4%	189,145	3.8%
Surgical Site Infection	24,445	7.9%	76,175	1.5%
Number of Infections					<.01
1	183,725	59.7%	326,795	6.5%
2	77,365	25.1%	95,060	1.9%
≥3	22,995	7.5%	122,00	0.2%
Type of infectious organism, n (%)					<.01
Gram Negative	13,445	4.4%	23,890	0.5%
Gram Positive	39,050	12.7%	62,935	1.3%
Other	159,435	51.8%	3,291,525	65.5%
Site of organ dysfunction, n (%)					<.01
Hematologic	59,370	19.3%	174,445	3.5%
Hepatic	5,105	1.7%	6,175	0.1%
Neurologic	73,740	24.0%	121,980	2.4%
Renal	123,355	40.1%	243,545	4.8%
Respiratory	122,250	39.7%	226,080	4.5%	<.01
LOS, mean (SE)	16	0.13	5	0.01	<.01
In-hospital Mortality, n (%)	29,145	9.5%	109,505	2.2%	<.01
Cost, median (IQR)	$29,438	$13,568–$68,515	$10,577	$6,245–$17,455

Elixhauser risk score for mortality; ISS = injury severity score; AIS = abbreviated injury score; LOS = length of stay

Multivariable analyses for health outcomes and cost of care between sepsis and non-sepsis trauma hospitalizations

Our survey-adjusted multivariable model included the following variables: age, sex, race/ethnicity, Elixhauser comorbidity score, insurance status, hospital region/type, bed size, injury severity score (severe vs. non-severe), and abbreviated injury score for each body region (serious vs. non-serious) (Table 2). An interaction between sepsis and severe injury (ISS ≥15) was present (p<0.01) in the association with mortality so stratified analyses by severe injury was performed. The risk factors most strongly associated with sepsis in the stratified multivariable analysis are reported in Table 2. Among the non-severe trauma group (ISS <15), injuries to the head/neck and extremities had stronger associations for sepsis than other body regions. Among the severe trauma subgroup (ISS ≥15), injuries to the chest and abdomen/pelvis were more strongly associated with sepsis than other body regions.

Table 2.

Risk Factors Associated with sepsis after trauma hospitalization stratified by injury severity score (N=5,329,714)

Patient and Hospital Characteristics	Severe Injury (ISS≥15)			Non-Severe Injury (ISS<15)
	OR for Sepsis	95 %	CI	OR for Sepsis	95 %	CI
Age, (per year)	0.99	0.99	0.99	1.01	1.01	1.01
Female	0.90	0.87	0.94	0.93	0.90	0.95
Elixhauser Comorbidity Score	1.46	1.45	1.48	1.54	1.53	1.55
Race/Ethnicity
Non-Hispanic White	referent
Non-Hispanic Black	1.13	1.06	1.19	1.01	0.96	1.06
Hispanic	1.07	1.00	1.14	1.04	0.98	1.09
Other	1.11	1.03	1.19	1.01	0.94	1.07
Insurance
Private	referent
Medicare	0.90	0.85	0.95	1.20	1.15	1.25
Medicaid	1.40	1.32	1.48	1.49	1.41	1.58
Self-pay	0.84	0.78	0.90	1.10	1.02	1.18
Hospital Region/Type
Urban teaching	referent
Rural	0.58	0.51	0.66	0.74	0.70	0.78
Urban nonteaching	0.76	0.72	0.80	0.87	0.84	0.90
Bed Size
Small/Medium	referent
Large	1.09	1.03	1.15	1.12	1.09	1.16
Injury by body region
Head/Neck	1.90	1.81	2.00	2.13	2.04	2.22
Face	1.57	1.30	1.91	1.96	1.09	3.53
Chest	2.82	2.68	2.96	1.71	1.63	1.80
Abdomen/pelvis	2.39	2.09	2.72	1.00	-	-
Extremities	1.66	1.57	1.75	1.36	1.30	1.41
External surface	1.15	0.75	1.76	2.09	1.41	3.09

ISS = injury severity score; Bed size = Bed size categories are based on hospital beds, and are specific to the hospital’s location and teaching status. Bed size assesses the number of short-term acute care beds set up and staffed in a hospital. Hospital information was obtained from the American Hospital Association Annual Survey of Hospitals (https://www.hcup-us.ahrq.gov/db/vars/hosp_bedsize/nisnote.jsp)

Severely and non-severely injured septic patients had an odds ratio for in-hospital mortality of 1.39 (95% CI: 1.31, 1.47) and 4.32 (95% CI: 4.06, 4.59) comparted to patients without sepsis, respectively (Supplemental 2). The adjusted marginal cost of hospitalization for sepsis trauma patients compared to non-septic trauma patients was $16,646 (95% CI: $16,294, $16,997), which was greater than the marginal cost of hospitalizations for trauma patients with severe injury (ISS ≥15) compared to non-severe injury ($8,639; 95% CI: $8,424, $8,855) (Supplemental 3). In trauma hospitalizations with sepsis, the following factors were associated with increasing costs of care: increasing number of organ dysfunctions and infections, mechanical ventilation greater than or equal to 96 hours, shock, and gram-negative infections (Supplemental 4).

DISCUSSION

In our study of more than 5 million adults discharged from US community hospitals with trauma, non-severe injuries with sepsis had a three-fold greater change in the odds risk of mortality than severe traumatic injuries with sepsis. The increase in costs of care for trauma hospitalizations with sepsis was greater than the increase in costs of care associated with severe injury. The proportion of deaths in the patients with sepsis after trauma during the 5-year study period did not change across case-identification methods. The lack of improvement in mortality and high cost for sepsis care after trauma suggest focused efforts are needed to better identify cases and provide early treatment.

Mortality in sepsis has decreased overall and clinical trials reflect that usual care with early detection and treatment is key.⁴ Our trauma cohorts of more than 300,000 cases demonstrated no change in mortality over a 5-year period. Recent studies show that subtypes carry different prognoses, and trauma is another subtype to consider.^{7, 8} Nationally, patients with trauma have less comorbidities and are younger than non-trauma patients with sepsis, which may contribute to differing outcomes.³³ Some similarities do exist to our study with the majority of infections comprising respiratory and genitourinary sources.³⁴ We showed demographics, comorbidities, insurance status, body region of injury, and hospital-level factors were associated with sepsis. Other reports also describe Glasgow Coma Scale, blood product transfusion, operative procedures, and need for laporatomy as factors associated with sepsis.^{11, 12, 35}

The interaction between sepsis and trauma severity in our study highlighted the different odds risk for death within the trauma cohort. The risk for death in the severely injured group was greater in those with injuries to the chest and abdomen/pelvis, which may have accounted for the organ dysfunctions and sites of infection that also occurred in greater proportions to these regions. In the less severely injured group, a greater odds risk for sepsis occurred in regions of the head/neck and extremities which has not been described previously. Risk stratification for sepsis by severity of injury and location may have implications for prevention or early treatment of sepsis.

Costs associated with sepsis were nearly two-fold higher than costs associated with severe injury. The complications from sepsis included organ dysfunctions and mechanical ventilation, which may have further contributed to cost. Treating sepsis does carry a high cost in postoperative cases,³⁶ but few studies describe the incremental cost associated with sepsis in patients after traumatic injury. Research towards improving care in sepsis cases after trauma are needed, especially since we showed the cost for sepsis care was greater than the cost for severe trauma.

Our observed case-rate and mortality in trauma hospitalizations with sepsis were lower than nationally representative and large-scale studies in mixed cohorts of sepsis.^{5, 37} To our knowledge, this is the first report on the epidemiology of sepsis after trauma using national estimates representative of all US community hospitals. Trauma centers in Pennsylvania reported an incidence of 2%⁹, similar to a study in the National Trauma Databank¹², but these studies were prior to 2009 and before widespread adoption of comprehensive electronic health records and mandates for quality performance reporting in sepsis care.³⁸ A study from the German trauma registry did have a case-rate for sepsis of 10.2% but in a sampling from less than a third of their trauma centers. Their inclusion criteria also required admission to the intensive care unit and an ISS ≥9.³⁵ The variability in our proportions was in part due to the higher sensitivity in using implicit codes and the higher specificity in using explicit codes.

New state and national sepsis quality measure reporting have affected coding behavior.³⁷ The rise in the incidence of hospitalizations with sepsis codes are in the wake of septicemia diagnoses increasingly being applied to patients without positive blood cultures.³⁹ Culture-negative sepsis preclude the use of trauma databases such as TQIP, which necessitate positive blood cultures to classify cases of sepsis. Nevertheless, the lack of a gold standard for sepsis identification contribute to the heterogeneity in reports of sepsis incidence and we demonstrated this problem continued in the trauma population between implicit and explicit sepsis codes.

Limitations to our study include the reliability of claims-based surveillance for sepsis cases. The results reported in this study are useful for establishing the epidemiology of trauma and sepsis but may not reflect true incidence. Rhee et al. demonstrate that although claims data indicate a rapid increase in the incidence of sepsis, clinical data show sepsis incidence remain stable.³⁹ Hospitals also vary in how they assign codes for infection and organ dysfunction, so the presence of these codes at discharge may fulfill the Angus implicit ICD codes criteria but may misrepresent sepsis cases, especially in traumas where organ dysfunction may be attributable to the trauma itself.¹¹ Lastly, while our study from NIS was representative of all community hospitals in the US, it was focused in only adults and in-hospital deaths and should not be generalized to children or mortality after discharge.

CONCLUSION

Sepsis is the leading cause of death for acute care hospitalizations, but little is known in the cohort with traumatic injury. We showed the case-rate of trauma after sepsis was highly variable based on existing case-definitions but trends for in-hospital mortality have not improved, regardless of the case definition. Both severe and non-severe injuries had an increased odds risk for death in cases with sepsis and the incremental cost for sepsis care was greater than the incremental cost for severe trauma. Dedicated effort to improve sepsis care in trauma populations are needed, starting with better methods to identify cases and targeted effort to reduce mortality.