Association of Hospital Altitude and Postoperative Infectious Complications After Major Operations

Introduction

The medical community is increasingly finding uses for hyperbaric hyperoxic states, including the treatment of severe infections and chronic wounds (Goldman, 2009; Kranke et al., 2015; Bennett et al., 2016). The inverse found at high altitude, hypobaric hypoxia, has been long studied in terms of physiologic changes (Coppel et al., 2015) and has even been linked to clinical outcomes such as increased postoperative thrombus formation (Damodar et al., 2018), decreased tuberculosis infection, morbidity, and mortality (Perez-Padilla and Franco-Marina, 2004). However, there has been no investigation of whether the benefits seen in hyperbaric hyperoxic states correspond to worsened infectious and regenerative outcomes in hypobaric and hypoxic states.

This is of particular concern for surgical patients as surgical site infections (SSI) and their associated cost, morbidity, and mortality have become major obstacles in providing cost-effective and quality health care. The American College of Surgeons (ACS) has recently reported that SSIs have become the most common hospital-acquired infection and are the costliest to treat (Ban et al., 2017). The overall rate of SSI for all inpatient surgery is 2%–5%, and each incidence costs ∼$20,000 (Ban et al., 2017). Given that SSIs are the most common cause for postoperative readmission in the United States, it is estimated that they cost the health care system $3.5–10 billion per year (Merkow et al., 2015). Due to the complex and multifactorial causes of SSIs, several multimodal protocols have been implemented with varying degrees of success (Deery et al., 2019). The ACS 2016 Update lists 30 risk factors for SSI, divided into intrinsic (patient-related) and extrinsic (procedure-related) categories, but does not consider potential environmental influences such as elevation (Ban et al., 2017).

Given the potential economic and clinical importance, we used ACS National Surgical Quality Improvement Program (NSQIP) data, in conjunction with Google Maps, to investigate the rates of key 30-day postoperative infectious outcomes, including urinary tract infection (UTI), SSI, and pneumonia in hospitals across the United States. We hypothesized that hospital elevation would correlate positively with postoperative infectious complication rates.

Materials and Methods

The 2016 ACS NSQIP data, used for this analysis, included 999,768 patient records from 680 hospitals. This dataset was first limited to U.S. hospitals, for which altitude was determinable by hospital zip code. This filtering resulted in 850,108 patient records from 579 hospitals, whose altitude ranged from 0 to 6206 feet above sea level. For each outcome, patient records were selected from NSQIP-targeted groups of current procedural terminology (CPT) codes that tended to exhibit the highest rates for the specific outcome. This resulted in an outcome-enriched database of 82,172 patients for analysis of SSI, 175,409 patients for analysis of UTI, and 88,856 patients for the analysis of pneumonia. The full procedure list can be seen in Table 1.

Altitude was determined for each hospital at which each procedure was performed. Zip codes for each hospital were provided by the ACS. They were individually looked up using Google Maps (San Jose, CA) by C.W., and the resulting list was returned to the M.E.C. who merged these results with the patient dataset.

We performed a sensitivity analysis using five different methods to study data: (1) As a preliminary analysis, raw rates were examined for seven altitude groups (0′–500′, 500′–1500′, 1500′–2500′, 2500′–3500′, 3500′–4500′, 4500′–5500′, and 5500′+), and the Cochran-Armitage test was used to ascertain the presence of linear trend; (2) a univariate logistic model was used to predict the patient-level binary outcome from the continuous altitude variable; (3) a univariate model was used to predict the patient-level binary outcome from the continuous altitude variable, and there was an adjustment in variance estimates to account for the clustering of patients within hospitals (without this adjustment, it is possible that data from higher altitudes—from few hospitals but still from many patients—will be treated as more reliable than they are); (4) a multivariate logistic model was used where the continuous altitude variable was forced into the model and then stepwise selection was used to add additional NSQIP patient-level variable; and (5) using the predictors included in the multivariate logistic model, there was an adjustment in variance estimates to account for the clustering of patients within hospitals. Odds ratios (ORs) were reported for all 12 (three outcomes by four analyses) logistic models, each of which used increments of 1000 feet. Thus, the ORs represent the change in risk for each 1000-foot increase in elevation.

For the multivariate models, NSQIP variables used for benchmarking were used to risk adjust the model so that effect of altitude would be estimated while controlling for any differences due to patients and procedures. To avoid overfitting, a forward selection process was used. Definitions are not reported here, but conform to NSQIP specifications. Of note, outcome-specific CPT linear risk is generated from a preliminary multiyear database and estimates the operation's risk at the granularity of the principle CPT code and is routinely the first variable selected in NSQIP risk-adjustment models. Although not all variables are selected, the nature of collinearity among predictors indicates that the selected set includes almost all the relevant information contained among the predictors.

Results

Overall patient demographics can be seen in Table 2. There was no evidence of a linear trend across all outcomes as shown in Table 3 (two-tailed p = 0.250, 0.184, 0.120, for SSI, UTI, pneumonia, respectively). Univariate modeling with or without clustering was not statistically significant for any outcome (Table 4). Multivariate analysis revealed no statistical significance for UTI (p = 0.402, 0.543 with and without clustering, respectively) or SSI (p = 0.853, 0.914 with and without clustering, respectively). Multivariate analysis revealed altitude had a statistically significant impact on postoperative pneumonia with an OR of 0.930 (p = 0.031) in the clustered model (Table 4). The variables in the forward selection multivariate model and order of selection are shown in Table 5.

Discussion

Postoperative complications increase the cost of care, burden the health care system, and most importantly negatively impact patient outcomes. We analyzed the relationship between hospital altitude and occurrence of infectious complications within 30 days of operations at higher risk of postoperative infections using several methods. While the multivariate modeling represents the most robust analysis, the consistency between all analysis types increases the strength of our findings. Similarity of results across the sensitivity analyses (Table 5) and the comprehensive nature of the risk adjustment suggests that confounding of altitude with characteristics of patient or procedure risk is not a tenable hypothesis.

We found no correlation between increasing altitude and increased risk of postoperative SSIs or UTIs. Hospitals, clinicians, and patients practicing in higher altitudes should be reassured by the findings indicating that their patients do not suffer worse operative outcomes due to their geographic predisposition. Importantly, future efforts to risk-adjust hospital and provider outcomes do not need to take into account altitude for SSI and UTI.

The finding of a decreased risk of postoperative pneumonia at higher altitudes was surprising. Using the reported OR, a hospital in Denver, CO, being roughly 5000 feet above a hospital near sea level, could experience a postoperative pneumonia risk of 0.695 (0.93⁵) compared to its lower counterpart. This is a significant difference and there exist several possible and clinically relevant explanations.

Exposure to oxygen allows for the formation of reactive oxygen species (ROS), leading to tissue damage and inflammation (Brueckl et al., 2006). Given this, it is possible that relatively higher oxygen at lower altitudes leads to increased ROS in the pulmonary parenchyma, creating a baseline inflamed state that is more susceptible to infection, or allowing for increased inflammation during an initial insult, such as a subclinical low respiratory infection, allowing it to progress to a full pneumonia. If true, this could indicate a benefit to ROS mitigation at lower altitudes.

Alternatively, it is possible that during adaptation to higher altitude, lung perfusion patterns change such that leukocytes are delivered at baseline to more lung regions, leading to a more rapid immune response to potential pathogens. Were this difference due to a physiologic adaptation to either hypoxia or even hypobaric pressures, we hazard it would be due to chronic adaptations, which are different than acute adjustments (Coppel et al., 2015), given it is likely that most database patients were operated on in their own state. However, this is unverified and warrants additional investigation. Given that the highest altitude contained in this study was at 6206 feet and that significant physiological changes generally occur at altitudes higher than this, we only cautiously assert this physiologic hypothesis. Recent work with mice has demonstrated chronic hypoxia alters leukocyte metabolism such that negative effects of acute hypoxia on infectious morbidity and mortality, are negated in preconditioned mice (Thompson et al., 2017). This favors a metabolic and chronic explanation of the discrepancy.

Looking outside of the patient, we hypothesize that potential alteration of pathogen physiology could lead to fewer infections. Specifically, the decreased partial pressure of oxygen at altitude could theoretically lead to a decreased proliferation of aerobic bacteria, lessening their pathogenicity.

It must be noted that the previous observations showing decreased mortality and morbidly of patients with tuberculosis at altitude have not revealed an immunologic or physiologic cause (Perez-Padilla and Franco-Marina, 2004; Eisen et al., 2013). It is possible that both observations share the same cause and serves to increase the importance of further inquiry.

Limitations of this study include the unknown altitude at which patients recovered after discharge, and the possibility of patients developing wound infections after discharge to a different altitude. Because this analysis does not differentiate between patients chronically acclimated versus acutely exposed to high altitude, relative hypoxia within an altitude may not be the same for all patients and represents a study weakness. The study is limited to hospitals participating in the ACS NSQIP during 2016. Strengths of this study are the large sample size of patients and the rigorous methodology to account for altitude and clustering by hospitals, which was made possible by data analysis performed by the ACS (MEC, XM)—hospital identifiers are not included in the standard ACS NSQIP Participant User File.

Given the massive national effort to reduce surgical complications and improve outcomes, the discrepancy in postoperative pneumonia rates warrants further investigation. Identification of a clear etiology could lead to the development and implementation of a clinical intervention for patients at lower altitudes. Given the increased focus on using hospital- and provider-specific surgical complication rates to guide care improvement, these findings also beg the question whether a higher standard for postoperative pneumonia should be expected from high altitude hospitals. Future studies should also examine whether these protective effects hold equally true for both patients acutely and chronically exposed to higher altitudes.