Effects of Hospital-Acquired Pneumonia on Long-Term Recovery and Hospital Resource Utilization Following Moderate to Severe Traumatic Brain Injury

Abstract

Background:

Individuals with Traumatic Brain Injury (TBI) have extended inpatient hospital stays that includes prolonged mechanical ventilation, increasing risk for infections, including pneumonia. Studies show the negative short-term effects of hospital-acquired pneumonia (HAP) on hospital-based outcomes; however, little is known of its long-term effects.

Methods:

Prospective cohort study. National Trauma Databank (NTDB) and Traumatic Brain Injury Model Systems (TBI-MS) were merged to derive a cohort of n=3717 adults with moderate-to-severe TBI. Exposure data were gathered from the NTDB, and outcomes were gathered from the TBI-MS. The primary outcome was the Glasgow Outcome Scale-Extended (GOS-E), which was collected at 1, 2 and 5 years post-injury. GOS-E was categorized as favorable (>5) or unfavorable (≤5) outcomes. A generalized estimating equation model was fitted estimating the effects of HAP on GOS-E over the first five years post-TBI, adjusting for age, race, ventilation status, brain injury severity, injury severity score (ISS), thoracic Abbreviated Injury Scale score ≥3, mechanism of injury, intraventricular hemorrhage, and subarachnoid hemorrhage.

Results:

Individuals with HAP had a 34% (OR=1.34, 95% CI 1.15, 1.56) increased odds for unfavorable GOS-E over the first five years post-TBI compared to individuals without HAP, after adjustment for covariates. There was a significant interaction between HAP and follow-up, such that the effect of HAP on GOS-E declined over time. Sensitivity analyses that weighted for non-response bias and adjusted for differences across trauma facilities did not appreciably change the results. Individuals with HAP spent 10.1 days longer in acute care and 4.8 days longer in inpatient rehabilitation, and had less efficient functional improvement during inpatient rehabilitation.

Conclusions:

Individuals with HAP during acute hospitalization have worse long-term prognosis and greater hospital utilization. Preventing HAP may be cost-effective and improve long-term recovery for individuals with TBI. Future studies should compare the effectiveness of different prophylaxis methods to prevent HAP.

Background:

An estimated 2.5 million Americans annually have an emergency department visit, are hospitalized with, or die due to traumatic brain injury (TBI).¹ Nearly half of patients hospitalized with TBI experience long-term morbidity, contributing to a large proportion of US citizens living with chronic disability from TBI.² To prevent the high disability burden and associated costs, TBI researchers have focused on identifying acute predictors of long-term disability, including injury severity based on the Glasgow Coma Scale (GCS),^3-5 systemic hypotension,⁶ intracranial pressure,⁷ and post-traumatic hydrocephalus^8,9 among others.

An underemphasized area of TBI research is the contribution of acute non-neurological complications and conditions to long-term recovery. Many patients with moderate-to-severe TBI require mechanical ventilation in the days following their injuries and are susceptible to infection. A common result of mechanical ventilation and susceptibility to infection is hospital-acquired pneumonia (HAP). In observational studies, HAP incidence rates range from 30–61% in TBI populations.^10-12 Variation in incidence estimates arise from heterogeneous, single-center cohorts, including cohorts restricted to only ventilated patients. Ventilation is an important predictor for incident HAP following TBI, with each additional ventilator day conferring a 7% increased risk for infection.¹³ Other variables associated with HAP incidence post-TBI include thoracic Abbreviated Injury Scale (AIS) score ≥3 and gastric aspiration.¹² Some individuals with TBI also experience a period of acute lymphocyte dysfunction following injury, known as lymphopenia,¹⁴ resulting in suppressed immunity and a decreased capacity to fight acquired infections, like pneumonia.^14-19

HAP development leads to direct increases in healthcare utilization and expenditures. Critically ill patients with HAP accrue roughly $40,000 more in acute hospitalization costs, and require approximately twice the intensive care unit and hospital lengths of stay (LOS) compared to critically ill patients without HAP.^10,20 One study evaluating acquired brain injury patients determined that individuals with ventilator-associated pneumonia (VAP) had higher hospital costs, longer LOS, and more readmissions compared to individuals without VAP matched on age, sex, diagnosis, date of admission, and hospital size.²¹

Past TBI studies characterizing HAP have examined short-term associations of HAP with greater cost and worse acute hospital outcomes.^10,13,22,23 A research gap exists in understanding the long-term associations with HAP in this population. Two recent small studies lend preliminary evidence that HAP may negatively impact long-term outcomes following injury.^11,24 To build upon these studies, a large multi-site longitudinal prospective study is necessary to estimate the long-term effects of HAP on disability and hospital resource utilization post-TBI. As an initial step, we leveraged a large probabilistically-merged database of the National Trauma Databank (NTDB) and TBI Model Systems (TBI-MS) National Database. Study objectives included: 1) determining the long-term effects of HAP on disability after moderate-to-severe TBI; and 2) comparing hospital resource utilization metrics between individuals with moderate-to-severe TBI, with/without HAP. We hypothesized that individuals with HAP have significantly poorer long-term outcomes over time, and have longer LOS during acute and rehabilitation inpatient care.

Methods:

All TBI Model Systems centers represented in this study had approved local Institutional Review Board protocols. Individuals (or their proxies when appropriate) signed informed consent to participate in data collection from the acute and rehabilitation phases of care as well as longitudinal follow-up with the general goal to learn more about TBI outcomes. We used data from two large databases: the NTDB and the TBI-MS National Database. The NTDB is the largest aggregation of trauma registry data in the United States. Participating hospitals contribute information on all trauma patients treated at their institution. Deidentified data are submitted to the NTDB and compiled for hospital benchmarking, data quality reports, and addressing trauma-related research questions. The TBI-MS is a prospective cohort study that includes data collected at up to 20 acute rehabilitation centers. Included patients received acute care within 72 hours of injury at a designated acute care hospital, survived through acute care, and were stable medically to receive rehabilitation. Other TBI-MS inclusion criteria include: a moderate-to-severe TBI (defined by at least one of the following: post-traumatic amnesia>24 hours, trauma-related intracranial neuroimaging abnormalities, loss of consciousness exceeding 30 minutes, or a GCS<13), age 16+ years at time of injury, and presentation to a TBI-MS acute care hospital within 72 hours of injury.²⁵ Data are collected at enrollment and follow-up interviews occur years 1, 2, and 5, as well as every 5 years afterward until death.

We used a probabilistic matching algorithm to combine the de-identified NTDB and TBI-MS. The merger of the two databases was possible because participants in the TBI-MS had a trauma record submitted to the NTDB. We developed the algorithm in two sites where exact matches on patient identifiers were available to form a deterministic dataset to quantitatively assess the sensitivity and positive predictive value of our algorithm. We previously published detailed methods used for algorithm development²⁶ and validation.²⁷ The final NTDB-TBI-MS cohort contained n=4022 individuals with TBI, injured between 1998–2015. The present study included individuals the TBI-MS National Database, who also were probabilistically merged with NTDB records to derive the current cohort. We further restricted our cohort to participants injured between 1998–2013 to examine five-year outcomes (n=3712). A flow diagram of participant inclusion to the final analytic cohort is shown in Figure 1. There were 21 NTDB trauma facilities across 17 TBI-MS centers represented in the present dataset.

Figure 1.

Flow Diagram of Participants in Analytic Sample. N=3712 individuals were in the final analytic cohort. Of those individuals, N=3149 were followed-up at year 1, N=3086 were followed up at year 2, and N=2979 were followed up at year 5 after injury.

The specific variables used in this study from the NTDB and TBI-MS, along with a detailed description of each variable, are provided in Table 1. The primary exposure was HAP; specifically, cases developed during the acute hospitalization. Individuals were considered to have HAP from either NTDB complication codes or diagnoses codes collected as a part of the TBI-MS. The primary outcome was Glasgow Outcome Scale-Extended (GOS-E) score²⁸, assessed at 1, 2, and 5 years post-injury. For the purposes of this analysis and direct comparison with prior studies, scores were dichotomized ≤5 (unfavorable outcomes) vs. >5 (favorable outcomes), as reported previously.¹¹ Secondary descriptive analyses were conducted to examine associations between HAP and variables related to hospital resource utilization: acute care LOS, rehabilitation LOS, and change in Functional Independence Measure (FIM) scores over time (FIM efficiency). FIM efficiency is the change in FIM score during rehabilitation divided by the rehabilitation LOS. For post-hoc analyses, complications other than HAP were extracted from the NTDB complications code to calculate a non-HAP complication burden score.

Table 1.

Description of measures

NTDB measures	Construct	Description of Measure
HAP	Primary exposure; Pneumonia in hospital	NTDB complication code: Documentation for HAP was defined as one of two criteria, either: Rales or dullness on a physical examination of the chest and any of the following: new onset purulent sputum, organism isolated from blood culture, or isolation of pathogen from specimen obtained by transtracheal aspirate or biopsy; OR positive chest radiographic exam and any of the following: new onset purulent sputum, organism from the blood, isolation of pathogen from specimen isolated from transtracheal aspirate or biopsy, isolation of virus or detection of viral antigen from respiratory secretions, diagnostic single antibody titer (IgM) or 4-fold increase in paired serum sample (IgG) for pathogen, or histopathologic evidence of pneumonia The complication occurs > 48 hours after hospital admission
ISS	Injury Severity-all body regions	An anatomic trauma severity scale derived from the AIS that quantifies overall injury severity across eight body regions Scores range from 1-6, with 1 being minor and 6 unsurvivable ISS is the sum of squares of the 3 most severely injured body regions
Non-head ISS	Injury Severity-all non-head body regions	The Non-head ISS was calculated using AIS region scores The sum of squares of the 3 most severe non-head body regions
Thoracic AIS	Thoracic injury severity	AIS thoracic region score Scores range from 1-6, with 1 being minor and 6 unsurvivable Individuals with no thoracic injuries were assigned a score of 0 Significant thoracic injury was classified as a thoracic AIS score ≥3
Ventilation Status	Ventilation	ICD-9 Procedure code range from 96.7-96.72
Ventilation days	Number of days on a ventilator	Number of days on ventilator support among individuals on a ventilator
TBIMS measures
HAP	HAP diagnosis code	ICD-9 Diagnosis codes for HAP: 482-486
Age	Age	Age at injury
Race	Race	Grouped as White, Black, or Other
Brain Injury Severity	Brain Injury Severity based on DoD scale	Calculated variable consisting of two categories, moderate or severe injuries, which is a function of the Glasgow Coma Scale (GCS), post-traumatic amnesia (PTA), and number of days it takes to follow commands Moderate injuries are defined by a GCS between 9-12, PTA between 0-7 days, or days to follow commands ≤1 day Severe injuries were defined as a GCS between 3-8, PTA >8 days, or days to follow commands >1 day
MOI	Cause of injury	Grouped as Motor Vehicle, Fall, Pedestrian, or Other
CT scan findings	Nature of TBI	Trained abstractors review patient electronic medical records to document presence or absence of neuropathology on computed tomography images Grouped as presence/absence of: SDH, EDH, IVH, and SAH
Payor Status	Insurance	Payor was dichotomized as Medicare or Medicaid vs. all other payment sources
Rehabilitation Interruption	Interruption to Rehabilitation Care	Short-term interruptions during inpatient rehabilitation to acute care
Acute care LOS	Number of hospital days in acute care	Hospital days in acute care
Rehabilitation LOS	Number of hospital days in rehabilitation	Hospital days in rehabilitation
FIM efficiency	Rate of change in FIM over the course of Rehabilitation	Calculated variable from the formula: (FIMdischarge-FIMadmission)/(Rehab LOS)
Rehospitalization	5- year Rehospitalization	All-cause rehospitalization at any point at 1, 2, or 5 years post-injury
GOS-E	Primary outcome; Global disability measure	A trained research assistant conducts a 20-question structured interview that answers a variety of questions pertaining to patient’s current condition, covering the following topics: 1) level of consciousness, 2) independence at home, 3) independence outside home, 4) return to work, 5) participation in social and leisure activities, 6) family and friendships, and their 7) return to normal life. The range of scores are from 1-8, with higher scores corresponding to better recovery. Scores were dichotomized ≤5 (unfavorable outcomes) vs. >5 (favorable outcomes)

Abbreviations: HAP, Hospital-acquired Pneumonia; ISS, Injury Severity Scale; AIS, Abbreviated Injury Scale; DoD, Department of Defense; MOI, Mechanism of Injury; LOS, length of stay, GCS, Glasgow Coma Scale; Post-traumatic Amnesia (PTA); Functional Independence Measure (FIM); GOS-E, Glasgow Outcome Scale-Extended Scale; CT, Computed Tomography; SDH, subdural hematoma, EDH, epidural hematoma, IVH, intraventricular hemorrhage; SAH, subarachnoid hemorrhage

Though probabilistic matching is common in birth cohorts and life-course studies,²⁹ its use in merging databases in trauma, surgery, and rehabilitation is novel. To evaluate the integrity of this approach for our purposes, we developed a deterministic cohort from two TBI-MS sites to conduct a post-hoc sensitivity analysis as an internal validation measure to examine the veracity of our primary findings from the probabilistically-derived cohort. This deterministic cohort included n=775 individuals with TBI injured between 2002–2013.

Statistical Analysis:

Demographic and clinical variables were compared by HAP status and GOS-E. Means and standard errors described continuous variables, and frequency and percentages described categorical variables. Chi-square tests compared categorical variables, and t-tests or Mann Whitney U test, were used to compare continuous variables by HAP status and GOS-E score.

A generalized estimating equation (GEE) model was used for the primary analysis assessing the relationship between HAP and 1, 2 and 5-year GOS-E. GEE models were fit using an unstructured correlation structure, binomial distribution, and logit link. We chose covariates that were associated with HAP and 1-year GOS-E at an α=0.20 level. Covariates included: age, race, ventilation status, brain injury severity, injury severity score (ISS), thoracic AIS ≥3, mechanism of injury, intraventricular hemorrhage status, and subarachnoid hemorrhage status. We adjusted for race in our analysis because of its documented associations with unfavorable outcomes in the literature³⁰ and in our study. The same covariates were included in the primary, secondary, and sensitivity analyses. To observe covariate effects on HAP associations with GOS-E, four models are presented: Model 1 (unadjusted), Model 2 (adjusts for age effects), Model 3 (adjusts for age and race only), and Model 4 (fully adjusted for all chosen demographic and clinical variables). A HAP*follow-up interaction was also tested in the primary cohort to test if the effect of HAP on GOS-E varies across follow-up period (1, 2, and 5 years). Primary results are reported with moderate and severe injury groups combined, and a HAP*injury severity interaction was tested.

Given our fixed sample size, an alpha=0.05, 1/3 rate of HAP, and 50% of non-HAP patients having unfavorable outcomes, we have 80% power to detect an OR of 1.24 for the unadjusted analyses. For multivariable analyses, if we assume the fully-adjusted covariate model explains as much as 50% of the variability in HAP occurrence we will still have 80% power to detect an OR of 1.35. Of note, power analyses were based on a logistic regression model, but we believe our sample size is sufficiently powered for a GEE model, which considers multivariable effects on binary outcomes over time.

In order to determine HAP effects on hospital resource utilization, analysis of covariance (ANCOVA) models were used to compute covariate-adjusted mean scores for acute LOS, rehabilitation LOS, and FIM efficiency by HAP status. SAS 9.4 was used for all statistical analysis (Cary, NC).³¹

A post-hoc analysis was tested using a HAP-associated complication score developed for this report. The score represents hospital complications that are significant associated with both HAP and GOS-E at 1 year. The percentage change in the beta of the HAP variable was noted before/after inclusion of the complication score to determine if this score confounded the relationship between HAP and GOS-E. We also assessed the post-hoc association between tracheostomy status, time to tracheostomy placement, and HAP status, as one potential modifiable procedure that could influence HAP risk. Specifically, we examined whether time to tracheostomy was associated with acute care LOS and FIM efficiency among HAP patients who were ventilated. We conducted a post-hoc analysis where we adjusted the primary model for trauma facility to assess whether the primary association was confounded by acute care trauma hospital. Finally, to address potential nonresponse selection bias, we weighted the primary model for factors associated with nonresponse for GOS-E measured at year 1, 2, and 5 years using inverse probability weighting methods,³² such that the IPW-model was based on a pseudo-population balanced on factors associated with follow-up at each timepoint.

Results:

Demographic and Clinical Variables by HAP Status:

The cohort included n=1212 (32.7%) individuals with HAP and n=2500 (67.3%) individuals without HAP. The demographic and clinical variables by HAP status are presented in Table 2. Individuals with HAP were more likely to be younger men when compared to individuals without HAP (p<0.001 both comparisons). ISS score and non-head ISS were significantly higher among individuals with HAP (p<0.001 both comparisons). A greater proportion of individuals with HAP had a thoracic AIS score ≥3, were more frequently on a ventilator, spent more days on a ventilator, and had longer acute and rehabilitation LOS compared to individuals without HAP (p<0.001 all comparisons). A greater proportion of individuals with HAP had an interruption during rehabilitation (p=0.027). A greater proportion of individuals with HAP had an intraventricular hemorrhage (IVH) and subarachnoid hemorrhage (SAH) compared to no HAP (p<0.01 both comparisons). Injury mechanism also varied by HAP status (p<0.001), with a greater proportion of MVC and a lower proportion of falls in the HAP group.

Table 2.

NTDB Demographics and Clinical Variables by HAP Status

Variable	HAP (n=1212)	No HAP (n=2500)	p-value
Age (mean, SD)	38.36 (16.56)	43.45 (20.52)	<0.001*
Sex (men, %¥)	970 (80.03)	1756 (70.27)	<0.001*
Race (n, %¥)			0.170
White	851 (70.21)	1678 (67.15)
Black	209 (17.24)	473 (18.93)
Other	152 (12.54)	348 (13.93)
Brain Injury Severity (n, %)			<0.001*
Moderate	77 (6.49)	696 (28.22)
Severe	1110 (93.51)	1770 (71.78)
ISS (median, IQR)	27 (20-35)	22 (16-29)	<0.001*
ISS Non-Head (median, IQR)	10 (1-18)	5 (1-13)	<0.001*
Thoracic AIS ≥3 (n,%)	477 (44.3)	603 (26.93)	<0.001*
Ventilation status (n, %¥)	795 (65.59)	972 (38.88)	<0.001*
Ventilation days (median, IQR)	8 (0-16)	0 (0-3)	<0.001*
Cranial surgery status (n, %¥)	279 (27.30)	551 (25.94)	0.418
Payor status (n, %¥)			0.517
Government assistance	420 (37.80)	896 (38.96)
Private pay	691 (62.20)	1404 (61.04)
Acute care length of stay (median, IQR)	26 (19-36.5)	14 (8-22)	<0.001*
Rehabilitation length of stay (median, IQR)	23 (15-36)	17 (11-28)	<0.001*
Rehabilitation Interruptions (n, %¥)	77 (6.63)	115 (4.84)	0.027*
CT Injury Type (n, %¥)
SDH	663 (55.02)	1296 (52.26)	0.115
EDH	139 (11.54)	311 (12.54)	0.382
IVH	405 (33.61)	651 (26.25)	<0.001*
SAH	860 (71.37)	1661 (66.98)	0.007*
Mechanism of Injury (n, %¥)			<0.001*
MVA	802 (66.23)	1240 (49.98)
Assault/Violence	90 (7.43)	270 (10.88)
Fall	211 (17.42)	697 (28.09)
Pedestrian	80 (6.61)	172 (6.93)
Other	28 (2.31)	102 (4.11)

Abbreviations: GCS, Glasgow Coma Scale; ISS, Injury Severity Score; SDH, Subdural hematoma; EDH, Epidural hematoma; IVH, intraventricular hemorrhage; SAH, subarachnoid hemorrhage

Demographic and Clinical Variables by GOS-E at 1 year:

Demographic and clinical variables by unfavorable/favorable 1-year GOS-E status are presented in Table 3. There were 1491 (47.5%) individuals with unfavorable GOS-E scores and n=1651 (52.5%) individuals with favorable GOS-E scores. There were significant differences in GOS-E by race and payor status and mechanism of injury (p<0.001 all comparisons). Individuals with unfavorable GOS-E scores were older, more often on a ventilator, had more ventilator days, longer acute care and rehabilitation LOS, and a greater proportion of interruptions in rehabilitation care compared to individuals with favorable GOS-E scores (p<0.001 all comparisons). Individuals with unfavorable outcomes more often experienced SDH, IVH, and SAH injuries (p<0.01 all comparisons).

Table 3.

NTDB Demographic and Clinical Variables by GOS-E Status at 1 Year

Variable	Unfavorable GOS-E (n=1491)	Favorable GOS-E (n=1651)	p-value
Age (mean, SD)	44.10 (19.32)	39.74 (19.41)	<0.001*
Sex (men, %¥)	1099 (73.71)	1213 (73.47)	0.880
Race (n, %¥)			<0.001*
White	953 (63.92)	1232 (74.62)
Black	344 (23.07)	221 (13.39)
Other	194 (13.01)	198 (11.99)
Brain Injury Severity (n, %)			<0.001*
Moderate	245 (16.91)	392 (23.80)
Severe	1204 (83.09)	1255 (76.20)
ISS (median, IQR)	25 (17-34)	25 (17-33)	0.182
ISS Non-Head (median, IQR)	8 (1-13)	8 (1-14)	0.675
Thoracic AIS ≥3 (n,%)	422 (31.24)	498 (33.56)	0.187
Ventilation status (n, %¥)	773 (51.84)	744 (45.06)	<0.001*
Ventilation days (median, IQR)	1 (0-12)	0 (0-6)	<0.001*
Cranial surgery status (n, %¥)	411 (32.01)	334 (22.58)	<0.001*
Payor status (n, %¥)			<0.001*
Government assistance	662 (48.53)	437 (28.41)
Private pay	702 (51.47)	1101 (71.59)
Acute care length of stay (median, IQR)	22 (13-33)	15 (9-23)	<0.001*
Rehabilitation length of stay (median, IQR)	23 (15-38)	16 (10-25)	<0.001*
Rehabilitation Interruptions (n, %¥)	117 (8.36)	48 (3.02)	<0.001*
CT Injury Type (n, %¥)
SDH	858 (58.13)	834 (50.73)	<0.001*
EDH	180 (12.20)	208 (12.65)	0.699
IVH	504 (34.15)	434 (26.40)	<0.001*
SAH	1057 (71.61)	1104 (67.15)	0.007*
Mechanism of Injury (n, %¥)			<0.001*
MVA	790 (53.27)	953 (58.00)
Assault/Violence	176 (6.76)	111 (6.76)
Fall	389 (26.23)	389 (26.23)
Pedestrian	94 (5.72)	94 (5.72)
Other	87 (5.30)	87 (5.30)

Abbreviations: GCS, Glasgow Coma Scale; ISS, Injury Severity Score; CT, Computed Tomography; SDH, Subdural hematoma; EDH, Epidural hematoma; IVH, intraventricular hemorrhage; SAH, subarachnoid hemorrhage

Primary analysis: GEE Model of GOS-E at 1, 2, and 5 years post-TBI using NTDB probabilistic cohort

The GEE models for the GOS-E primary analysis are provided in Table 4. In the unadjusted model, individuals with HAP had a 28% increased odds for unfavorable GOS-E scores compared to individuals without HAP (OR=1.28, 95% CI (1.14, 1.45), p<0.001). After adjustment for age only, individuals with HAP had a 48% increased odds for unfavorable GOS-E scores compared to individuals without HAP (OR=1.48, 95% CI (1.29, 1.69), p<0.001). After adding race, individuals with HAP had a 51% increased odds for unfavorable GOS-E scores compared to individuals without HAP (OR=1.51, 95% CI (1.32, 1.73), p<0.001). In the fully adjusted model, individuals with HAP had a 34% increased odds for unfavorable GOS-E scores compared to individuals without HAP (OR=1.34, 95% CI (1.15, 1.56), p<0.001). The interaction between HAP and follow-up period on GOS-E over time was significant (p=0.018), indicating that the effect of HAP on GOS-E tends to decrease each follow-up period over the first five years post-injury. The interaction term between HAP*injury severity was tested, but was not significant (p=0.728); therefore, the interaction term was dropped from the model. The estimates for the full model and all covariates are provided in Supplemental Table 1.

Table 4.

Probabilistic Cohort GEE Regression Model for Repeated Measures GOS-E at 1, 2, and 5 years

	Model 1£		Model 2§		Model 3¥		Model 4€
	Odds ratio (95% CI)	p-value	Odds ratio (95% CI)	p-value	Odds ratio (95% CI)	p-value	Odds ratio (95% CI)	p-value
HAP		<0.001*		<0.001*		<0.001*		<0.001*
No (reference)	(reference)		(reference)		(reference)		(reference)
Yes	1.28 (1.14, 1.45)		1.48 (1.29, 1.69)		1.51 (1.32, 1.73)		1.34 (1.15, 1.56)

^*:statistically significant at p<0.05

^£:Model 1: Unadjusted

^§:Model 2: Adjusted for Age only

^¥:Model 3: Adjusted for Demographic Variables: Age and Race

^€:Model 4: Adjusted Demographic and Clinical Variables: Age, Race, ventilation status, brain injury severity, ISS, Thoracic AIS ≥3, mechanism of injury, intraventricular hemorrhage status, subarachnoid hemorrhage status

Sensitivity analysis: GEE Model of GOS-E at 1, 2, and 5 years post-TBI using Deterministic cohort

Using only the deterministic cohort, the unadjusted analysis showed individuals with HAP were at a 44% increased odds for unfavorable GOS-E scores compared to individuals without HAP (OR=1.44, 95% CI (1.05, 1.98), p=0.022) (Table 5). After adjusting for age, individuals with HAP had a 69% increased odds for unfavorable GOS-E (OR=1.69, 95% CI (1.22, 1.34), p=0.002). After adding race, individuals with HAP had a 72% increased odds for unfavorable GOS-E (OR=1.72, 95% CI (1.23, 2.39), p=0.001). In the full adjusted model, individuals with HAP had a 63% increased odds for unfavorable GOS-E compared with individuals without HAP (OR=1.63, 95% CI (1.16, 2.30), p=0.005).

Table 5.

Deterministic Cohort GEE Regression Model for Repeated Measures GOS-E at 1, 2, and 5 years

	Model 1£		Model 2§		Model 3¥		Model 4€
	Odds ratio (95% CI)	p-value	Odds ratio (95% CI)	p-value	Odds ratio (95% CI)	p-value	Odds ratio (95% CI)	p-value
HAP		0.022*		0.002*		0.001*		0.005*
No (reference)	(reference)		(reference)		(reference)		(reference)
Yes	1.44 (1.05, 1.98)		1.69 (1.22, 1.34)		1.72 (1.23, 2.39)		1.63 (1.16, 2.30)

^*:statistically significant at p<0.05

^£:Model 1: Unadjusted

^§:Model 2: Adjusted for Age only

^¥:Model 3: Adjusted for Demographic Variables: Age and Race

^€:Model 4: Adjusted Demographic and Clinical Variables: Age, Race, ventilation status, brain injury severity, ISS, Thoracic AIS ≥3, mechanism of injury, intraventricular hemorrhage status, subarachnoid hemorrhage status

Secondary analysis: Hospital utilization variables by HAP status

After covariate adjustment, individuals with HAP, had on average 10.1 more days in acute care LOS and 4.8 more days in rehabilitation LOS (p<0.001 both comparisons), which negatively impacted FIM efficiency for those with HAP. The unadjusted FIM efficiency for individuals with HAP was 2.02 compared to 2.31 for individuals without HAP. After adjustment for covariates, individuals with HAP, had 0.29 reduced FIM efficiency during rehabilitation (p<0.001).

Post-hoc Analyses:

As a post-hoc analysis, a HAP-associated complication score was added as a covariate in the primary model, and the beta estimate for the HAP variable did not significantly change (% change β_HAP=1.028%); therefore, this burden score was dropped as a covariate. We also descriptively observed a 44.7% rate of tracheostomy in the HAP group and 27.9% in the no HAP group (χ²=84.8, p=<0.001). When examining timing of tracheostomy placement, time to tracheostomy was not associated with HAP. Among individuals with HAP, greater time to tracheostomy placement was associated with a longer acute care LOS, but not associated with FIM efficiency. The association between HAP and outcome remained significant after adjustment for trauma facility (aOR=1.45; 95% CI: 1.24, 1.70, p<0.001). Finally, there were n=563 with missing follow-up data at year 1, n=626 with missing follow-up data at year 2, and n=733 with missing follow-up at year 5. Demographic and injury characteristics associated with loss to follow-up are provided in Supplemental Tables 2-4. When we applied the IPW to the primary model (Supplemental Tables 5), the relationship between HAP and GOS-E did not appreciably change (aOR=1.34, 95% CI: 1.14, 1.57).

Discussion:

Our findings highlight HAP as a potential early modifiable risk factor impacting TBI recovery and hospital resource utilization. We observed a HAP incidence rate of 32.7%, which is similar to past estimates.¹¹ The present study provides evidence that HAP effects may extend beyond the infection period itself and persist for years post-injury, perhaps through the propagation of a chronic inflammatory milieu³³ and decreased or delayed rehabilitation gains. In addition to poorer long-term prognosis, individuals with HAP had longer LOS and decreased efficiency in attaining functional rehabilitation gains compared to those without HAP.

Recent work by Esnault¹² examined the effects of early-onset VAP among 175 individuals with severe TBI and observed an elevated odds (OR=2.71, 95% CI: 1.01–7.25) for more unfavorable GOS scores at 1-year. Similarly, a small pilot study (n=141) led by our research group¹¹ used a similar design as the present study. To avoid duplicity in reporting, no individuals in our prior study were included in our present analysis. We previously showed HAP carried a 4.6 times (95% CI: 1.80–11.60) increased odds for unfavorable outcomes in a longitudinal model.¹¹ The small sample size and single-site design by Esnault et al.¹² and Kesinger et al.¹¹ likely account for the inflated effect sizes and wide confidence intervals compared to the present larger study. Importantly, we also observed that HAP effects attenuate modestly (~5%) over the 5-year time period. Additionally, we observed that HAP effects were similar when GOS-E was analyzed as a multinomial variable (data not shown).

Previously, Zygun¹⁰ showed a VAP incidence rate of 45% among 134 individuals with severe TBI, which the authors report is nearly three times the rate reported for all general trauma patients, according to the National Nosocomial Infections Surveillance program report.³⁴ This observation supports the possibility that individuals with severe TBI may be particularly vulnerable to hospital infections. Some individuals with TBI experience a period of lymphopenia beginning early post-injury.¹⁴ Persistent lymphopenia, which has been documented in trauma populations^15-19 and shown with preliminary observations in TBI populations,¹⁴ may reflect a decreased capacity to fight exposure to pathogens. Clinical and injury factors, like prolonged mechanical ventilation, concurrent polytrauma and specifically thoracic injuries, and pulmonary aspiration also are HAP risk factors. More severely injured patients are at greater risk for developing both HAP and poor outcomes. HAP was associated with GOS-E in unadjusted models indicating that it is a strong marker for poor outcomes. Moreover, adjustment for several potential confounders, including injury severity, extra-cerebral injury severity, and mechanism of injury, shows that the harmful effects of HAP on GOS-E are still evident after adjusting for standard risk covariates. In addition, our post-hoc analysis adjusting for the HAP-associated complication score, provided some evidence that HAP remains significant to long-term outcomes even after adjusted for other hospital complications. However, we cannot rule out other unmeasured confounders that could impact the effect of HAP on GOS-E.

In our study, we documented a modest negative confounding of the effect of HAP on outcome driven by age. Younger age was associated with a greater incidence of HAP, but is protective against unfavorable outcomes post-TBI. Though this association may seem counterintuitive, the negative association between age and HAP has been previously documented in the TBI-MS³⁵ and aligns with the mechanism of injuries experienced by older vs. younger individuals. Recent epidemiological data have observed that older individuals more often sustain isolated brain injuries from falls,³⁶ whereas, younger individuals are more likely to suffer polytrauma injuries resulting from motor vehicle accidents.³⁶ These findings highlight that HAP prevention efforts should be directed across the age span, particularly among younger ages, where HAP is more common. Though, it is important to consider that because the present cohort is restricted to those individuals with TBI who survived their initial injury and also received inpatient rehabilitation, the observed negative relationship between age and HAP may be subject to some survival selection bias. Specifically, older individuals, particularly those with pre-injury anticoagulation, are more likely to die from their injuries compared to those with similar age/injury type who are not on anticoagulants.^37-40

HAP occurrence after TBI has significant implications for hospital resource utilization. Our results showed that individuals with HAP had longer acute and rehabilitation LOS compared to individuals without HAP. This may not be a causal relationship between HAP and LOS. Rather, there may be a bidirectional relationship between LOS and HAP wherein individuals with longer hospitalizations more often develop HAP, and HAP leads to longer LOS. Future studies should prospectively and systematically collect time of infection to better understand how a longer LOS influences HAP risk. We also determined that HAP is associated with rehabilitation interruptions. Also, individuals with HAP had decreased FIM efficiency by a factor of 0.29. In other words, patients with HAP require roughly 30% more days in rehabilitation to achieve similar functional gains as patients without HAP. This finding suggests that preventing HAP after TBI may yield substantial cost-related benefits.⁴¹

This study highlights the potential importance of infection prophylaxis in TBI populations; HAP is extremely common and often preventable. Given the high morbidity and mortality associated with TBI across the lifespan, identifying potentially modifiable factors that can result in improved outcomes cannot be underemphasized. The recent 2016 4^th edition Guidelines for the Management of Severe Traumatic Brain Injury⁴² recommends Level IIA evidence for early tracheostomy as a method for infection prophylaxis. In our study we observed a strong association between tracheostomy and HAP, however, did not observe an association between time to tracheostomy with HAP incidence in our cohort. Others have also suggested early extubation is important.⁴³ A potentially cost-effective intervention is early mobility protocols during acute care, for those who are able, which has been shown to be efficacious in reducing incidence of HAP in neurointensive care unit patients through the increased clearance of secretions.⁴⁴ Work by Mendez-Tellez and Needham⁴⁵ shows that physical rehabilitation for ventilated patients in particular has positive impact on ICU-acquired complications and functional status at hospital discharge.Another study in an acute stroke population found that early dysphagia screening and treatment was associated with decreased HAP rates.⁴⁶ Another study found an association between no functional oral intake and HAP incidence after TBI.⁴⁷ Chlorhexidine oral disinfectants have also been documented as effective for HAP prophylaxis.⁴⁸ Additionally, targeted temperature management and conjugate short-term antibiotic therapy showed some evidence for prevention of HAP in populations with cardiac arrest,⁴⁹ but needs to be evaluated further in TBI populations.

HAP presents a significant public health problem in TBI populations. Despite some promising interventions that have shown efficacy, there may be issues with clinical implementation such as non-compliance, non-modifiable patient factors, and insufficient institutional priorities placed on pneumonia prevention. Future TBI studies would benefit from conducting pragmatic trials for HAP prevention in real-world settings.

There are limitations of this work that warrant consideration. We assessed HAP instead of VAP because there was no clear information on time until infection in the NTDB. Despite this, our results indicate that HAP is a significant predictor of GOS-E, despite adjusting for ventilation status and other factors shown to contribute to both HAP and GOS-E scores. Furthermore, because timing of infection was not readily available, exposed cases included those with both early- and late-onset HAP. The etiology of early vs. late-onset HAP is likely different, with the former commonly the result of aspiration pneumonia, and the latter potentially related to prolonged ventilation. Future studies would be useful in distinguishing temporal dynamics of HAP etiology and onset during acute care hospitalization. Additionally, NTDB complication codes are subject to underreporting, which could result in an underestimation of HAP incidence and bias the effect towards the null. The NTDB definition of HAP does not distinguish what set of criteria were met for individuals to be classified as cases. Yet the operational definition of HAP, set by the NTDB, could directly influence the observed incidence rate. However, as documented by prior studies, the NTDB definition of pneumonia includes a more liberal criteria of positive chest x-ray and purulent sputum or an abnormal physical exam and positive blood culture. The two avenues of positive diagnosis result in greater sensitivity, but lower specificity, of disease classification, compared to a disease definition of only a culture-based criteria.⁵⁰ To allay some concerns of underreporting we used all available data sources in this study for our HAP exposure classification. Specifically, through the availability of a second data source we were able to supplement HAP exposure data using diagnosis codes from the TBI-MS National Database. The primary analysis was conducted using a probabilistically-matched cohort; therefore, a small percentage of the pairs may be incorrectly matched. However, based on our previous algorithm development²⁶ and validation²⁷ studies, we have empirical evidence to suggest mismatched pairs were likely very rare (<2%). Also, the converging results in our sensitivity analysis from a deterministic dataset support the observation that HAP predicts poor outcomes post-TBI observed in the primary analysis. In our secondary analysis, the data we present is a proxy for hospital resource utilization; however, we do not have claims or cost data that may be a truer measure of utilization. Due to the lack of medication information in the NTDB, we were not able to adjust for potentially important confounding by pharmacological interventions, such as anti-coagulants. We did not account for changes in resuscitation approaches over time, such as crystalloid therapy and component therapy that may lead to some temporal bias. Though the hospital complications by definition occur 48+ hours after admission, it is possible that the findings could be confounded by some community-acquired cases of pneumonia, particularly depending on intubation location (e.g. in the field or ED). Finally, due to the longitudinal nature of the study, there was some degree of lost to follow-up. In our post-hoc analysis when we applied IPW for nonresponse, the association between HAP and GOS-E did not meaningfully change, which suggests that though there are factors associated with follow-up, they are not resulting in major selection bias for the primary effect of HAP on GOS-E.

We demonstrate that HAP increases odds for unfavorable outcomes by 34% up to five years post-TBI. This study provides a meaningful contribution to the field by highlighting the deleterious association between acute care HAP and long-term outcome in a large sample of individuals with TBI. The work supports the need for future studies to expand prospective research on infection prophylaxis during acute hospitalization. TBI populations are particularly vulnerable to incident HAP, and concerted efforts are needed to assess how primary infection prevention might improve long-term recovery.