Annals of the American Thoracic Society

To the Editor:

Hospitals geographically localize clinically similar patients into wards to provide specialized care that improves patient outcomes and care, and lowers costs. When these wards exceed capacity, patients become “geographically dispersed” to alternate locations. For example, critically ill patients may “board” in emergency departments or alternate intensive care units (ICUs) when the clinically appropriate ICUs are at capacity. Such geographic dispersion has been demonstrated to be associated with increased hospital length of stay (LOS), ICU and in-hospital mortality, and incidence of hospital-acquired infections, in addition to lower adherence to quality metrics (15). In several single-center studies and a meta-analysis across the United States and Canada, geographic dispersion among general internal medicine ward patients was associated with increased in-hospital mortality and resident burnout (6), and among a mixed medical–surgical ward population with increased hospital LOS (7).

Geographic localization is particularly pertinent for pulmonary service patients, who may benefit from specialized care focused on their unique needs, such as frequent assessments by respiratory therapists. For example, within the University of Pennsylvania Health System (UPHS), the pulmonary services are staffed by a board-certified pulmonologist and internal medicine residents. The pulmonary wards have higher respiratory therapist–to-patient ratios; and oxygen, noninvasive ventilation, and high-flow nasal cannula capabilities that other wards do not.

However, variations in outcomes among pulmonary service patients who receive care on pulmonary wards as compared with those who are geographically dispersed have never been evaluated. Therefore, our objective was to determine if geographic dispersion is associated with adverse outcomes among hospitalized pulmonary service patients. We hypothesized that geographic dispersion would be associated with increased hospital LOS.

Preliminary results of this work were presented as an abstract at the American Thoracic Society International Conference 2019 (8).


We performed a retrospective cohort study of all pulmonary service patients admitted to wards from 2014 and 2015 within three UPHS acute care hospitals. Patients are admitted to the pulmonary service through several mechanisms: 1) direct admission from a UPHS outpatient pulmonary practice; 2) through the hospital emergency department, when the patient presents with a primary respiratory issue and is followed in an outpatient pulmonary practice of UPHS; 3) through interhospital transfer of a patient with a primary respiratory issue followed in a UPHS pulmonary practice; and 4) from the hospital’s medical ICU if the patient has high respiratory care needs (e.g., new tracheostomy, secretions requiring frequent suctioning, and need for noninvasive ventilation or high-flow nasal cannula), as determined by the ICU team. If a bed is available on the pulmonary ward at each hospital, the pulmonary service patient is preferentially admitted there. If not, the pulmonary service patient will be admitted to another floor, with similar patient-to-nurse ratios and telemetry capabilities, but possibly higher patient-to–respiratory therapist ratios and different oxygen delivery capabilities.

We excluded patients transferred between geographically localized and dispersed wards (n = 33 [2%]). The primary exposure was “geographic dispersion,” defined as admission to a nonpulmonary ward, for the entirety of hospitalization, excluding any period of ICU admission. The primary outcome was the rate of discharge (i.e., a proxy for hospital LOS). Secondary outcomes included in-hospital mortality, discharge to home, discharge to skilled nursing facilities (SNFs), and 90-day hospital readmissions, as these represent patient- and surrogate-centered outcomes (9).

We first performed unadjusted analyses of geographic dispersion with each outcome separately. We then estimated separate multivariable regression models, adjusting for covariates selected for inclusion a priori based on a conceptual model of geographic dispersion and prior research of inpatients, including age, gender, race, ethnicity, insurance type, Elixhauser comorbidity scores, the Centers for Medicare and Medicaid Services four-level severity risk adjustment (based on admitting diagnosis, demographics, and comorbidities) (10, 11), admission diagnosis category (from International Classification of Diseases-9 and -10 codes), daily mean number of pulmonary service patients by hospital during the hospitalization, code status (full vs. any limitation on life-sustaining therapies), hospital admission source, ICU as the ward admission source, and season (6, 7). Because in-hospital mortality is a competing risk for being discharged alive, we analyzed hospital LOS by comparing rate of discharge using a subdistribution hazard ratio (SHR) estimated from a Fine and Gray model that treated death as a competing risk (12). We performed generalized linear models with log links for the secondary outcomes. We included hospital as a fixed effect in unadjusted and adjusted analyses to account for clustering. We calculated E values for all of the adjusted models to determine the minimum strength of an association on the SHR or risk ratio (RR) that an unmeasured confounder would need to have with both the exposure and outcome, conditional on the measured covariates, to nullify the observed exposure–outcome association for the statistically significant findings, or to change the null findings by 20% (13, 14). The University of Pennsylvania Institutional Review Board approved this protocol as exempt.


The study population included 1,603 patients in 31 wards across 3 hospitals (Table 1). Median age was 60 years (interquartile range = 48–70), 51% were female, and 53% were of white race. The 31 wards had a median 32 operational beds (interquartile range = 30–43). A total of 928 (58%) patients were geographically dispersed.

Table 1. Patient characteristics and unadjusted outcomes*

VariableTotal (n = 1,603)Localized (n = 675)Dispersed (n = 928)
Patient characteristics   
  1991 (62)421 (62)570 (61)
  2332 (21)130 (19)202 (22)
  3280 (18)124 (18)156 (17)
 Age, median (IQR), yr60 (48–70)60 (48–68)61 (50–71)
 Female819 (51)347 (51)472 (51)
  White846 (53)343 (51)503 (54)
  Black657 (41)289 (43)368 (40)
  Other100 (6)43 (6)57 (6)
 Latino/Hispanic52 (3)29 (4)23 (3)
  Private705 (44)306 (45)399 (43)
  Medicare726 (45)300 (44)426 (46)
  Medicaid87 (5)36 (5)51 (6)
  Other85 (5)33 (5)52 (6)
 Elixhauser comorbidity score, median (IQR)4 (3–6)4 (2–6)4 (3–6)
 Centers for Medicare and Medicaid Services severity risk adjustment   
  Mild71 (4)36 (5)35 (4)
  Moderate335 (21)177 (26)158 (17)
  Major570 (36)255 (38)315 (34)
  Severe627 (39)207 (31)420 (45)
 Admission diagnosis category   
  Symptoms, signs, and abnormal clinical and laboratory findings, not elsewhere classified575 (36)248 (37)327 (35)
  Diseases of the respiratory system362 (23)182 (27)180 (19)
  Diseases of the circulatory system126 (8)42 (6)84 (9)
  Infectious diseases107 (7)31 (5)76 (8)
  Endocrine, nutritional, and metabolic diseases100 (6)46 (7)54 (6)
  Other§333 (21)126 (19)207 (22)
 Daily mean number of pulmonary service patients by hospital during the hospitalization, median (IQR)21 (7–27)22 (8–28)20 (7–26)
 Code status—full1,266 (79)581 (86)685 (74)
 Hospital admission source   
  Emergency department1,231 (77)544 (81)687 (74)
  Outside facility280 (18)100 (15)180 (19)
  Direct92 (6)31 (5)61 (7)
 Ward admission source—intensive care unit712 (44)277 (41)435 (47)
  Winter407 (25)172 (26)235 (25)
  Spring458 (29)180 (27)278 (30)
  Summer429 (27)169 (25)260 (28)
  Fall309 (19)154 (23)155 (17)
Outcomes (all patients)   
 Hospital length of stay, median (IQR), days7 (4–14)6 (3–10)8 (4–16)
 In-hospital mortality151 (9)30 (4)121 (13)
Outcomes (survivors)n = 1,452n = 645n = 807
 Discharge destination   
  Home1,088 (75)521 (81)567 (70)
  Skilled nursing facility207 (14)66 (10)141 (17)
  Hospice78 (5)25 (4)53 (7)
  Rehabilitation facility39 (3)16 (2)23 (3)
  Transferred to another acute care hospital17 (1)10 (2)7 (0.1)
  Other||23 (2)7 (1)16 (2)
 90-d hospital readmissions460 (32)208 (32)252 (31)

Definition of abbreviation: IQR = interquartile range.

*All values presented as n (%) except where otherwise specified.

Other includes: Asian; American Indian/Alaskan Native; Pacific Islander; mixed; unknown; other.

Other includes: workman’s compensation; self-pay; charity; enrolled in a study; hospice; veteran’s association; missing.

§Other includes: neoplasms; diseases of the blood and blood-forming organs and certain disorders involving the immune mechanism; mental, behavioral, and neurodevelopmental disorders; diseases of the nervous system; diseases of the digestive system; diseases of the genitourinary system; pregnancy, childbirth, and the puerperium; diseases of the skin and subcutaneous tissue; diseases of the musculoskeletal system and connective tissue; congenital malformations, deformations, and chromosomal abnormalities; injury, poisoning, and certain other consequences of external causes; factors influencing health status and contact with health services; multiple diagnoses. Each of these diagnoses represented <5% of the overall population, and was derived from International Classification of Diseases-9 and -10 codes.

||Other includes: correctional facility; psychiatric facility; left against medical advice; user-defined code.

In unadjusted analyses, geographic dispersion was associated with decreased rates of discharge, increased in-hospital mortality, decreased discharge to home, and increased discharge to SNFs. After adjustment for potential confounders, the SHR estimated from a Fine and Gray competing risks model was 0.8 (95% confidence interval [CI] = 0.7–0.9, P = 0.001, E = 1.3), indicating that the rate of discharge was lower for dispersed patients (suggesting longer adjusted LOS). Geographically dispersed patients had: no difference in in-hospital mortality (RR = 1.2, 95% CI = 0.9–1.6, P = 0.3, adjusted percent: localized, 8% vs. dispersed, 10%, E = 1.6); lower rate of discharge home (RR = 0.95, 95% CI = 0.91–0.99, P = 0.03, adjusted percent: localized, 77% vs. dispersed, 73%, E = 1.3); higher rate of discharge to SNFs (RR = 1.5, 95% CI = 1.1–2.0, P = 0.01, adjusted percent: localized 11% vs. dispersed 16%, E = 2.3); and no difference in 90-day hospital readmissions (RR = 1.0, 95% CI = 0.8–1.1, P = 0.6, adjusted percent: localized 33% vs. dispersed 31%, E = 1.8) (Table 2).

Table 2. Associations of geographic dispersion with patient outcomes

Estimate (95% CI)P ValueEstimate (95% CI)P Value%
Primary outcome—subdistribution hazard ratio     
 Rate of discharge0.6 (0.6–0.7)<0.0010.8 (0.7–0.9)0.001
Secondary outcomes—risk ratios     
 In-hospital mortality3.5 (2.3–5.3)<0.0011.2 (0.9–1.6)0.3Localized: 8
     Dispersed: 10
 Discharge to home0.9 (0.8–0.9)<0.0010.95 (0.91–0.99)0.03Localized: 77
     Dispersed: 73
 Discharge to skilled nursing facilities1.7 (1.3–2.2)<0.0011.5 (1.1–2.0)0.01Localized: 11
     Dispersed: 16
 90-d hospital readmissions1.0 (0.8–1.1)0.71.0 (0.8–1.1)0.6Localized: 33
     Dispersed: 31

Definition of abbreviation: CI = confidence interval.

*Models were adjusted for: age, gender, race, ethnicity, insurance type, Elixhauser comorbidity scores, the Centers for Medicare and Medicaid Services four-level severity risk adjustment (based on admitting diagnosis, demographics, and comorbidities), admission diagnosis category (from International Classification of Diseases-9 and -10 codes), daily mean number of pulmonary service patients by hospital during the hospitalization, code status (full vs. any limitation on life-sustaining therapies), hospital admission source, intensive care unit as the ward admission source, season, and the patient’s hospital.

Estimated using a Fine and Gray competing risks model to account for the competing risk of death.


This is the first study to evaluate the association of geographic dispersion with adverse outcomes among ward-based pulmonary service patients. We demonstrated that geographic dispersion is independently associated with decreased rates of discharge (suggesting longer adjusted LOS) and discharge home, and increased rate of discharge to SNFs. Several mechanisms may account for these findings: geographically dispersed patients may have decreased access to respiratory therapists and pulmonary therapies; case managers on dispersed wards may not be aware of home services required for pulmonary patients; and primary teams may be less available for direct bedside care.

This study has important limitations. First, as a single–health system study, the results may not be broadly generalizable. Furthermore, many hospitals do not have subspecialty services. Having these services could create trade-offs between expertise and flexibility, which have unknown effects on hospital bed flow. Second, although we performed analyses adjusted for patient, ward, and hospital factors, there may be residual confounding. Specifically, dispersion may have been considered appropriate for patients in transition to comfort-focused care, leading to potential confounding by indication. Finally, the study may have been underpowered to detect differences in in-hospital mortality.

In summary, geographic dispersion among hospitalized pulmonary service patients is associated with decreased rates of discharge and discharge to home, and increased rate of discharge to SNFs. Future work is needed to confirm these findings in other settings and to understand the mechanisms underlying these differences.

1 . Chalfin DB, Trzeciak S, Likourezos A, Baumann BM, Dellinger RP; DELAY-ED Study Group. Impact of delayed transfer of critically ill patients from the emergency department to the intensive care unit. Crit Care Med 2007;35:14771483.
2 . Vishnupriya K, Falade O, Workneh A, Chandolu S, Landis R, Elizabeth K, et al. Does sepsis treatment differ between primary and overflow intensive care units? J Hosp Med 2012;7:600605.
3 . Pascual JL, Blank NW, Holena DN, Robertson MP, Diop M, Allen SR, et al. There’s no place like home: boarding surgical ICU patients in other ICUs and the effect of distances from the home unit. J Trauma Acute Care Surg 2014;76:10961102.
4 . Lott JP, Iwashyna TJ, Christie JD, Asch DA, Kramer AA, Kahn JM. Critical illness outcomes in specialty versus general intensive care units. Am J Respir Crit Care Med 2009;179:676683.
5 . Singer AJ, Thode HC Jr, Viccellio P, Pines JM. The association between length of emergency department boarding and mortality. Acad Emerg Med 2011;18:13241329.
6 . Thanarajasingam U, McDonald FS, Halvorsen AJ, Naessens JM, Cabanela RL, Johnson MG, et al. Service census caps and unit-based admissions: resident workload, conference attendance, duty hour compliance, and patient safety. Mayo Clin Proc 2012;87:320327.
7 . Song H, Tucker AL, Graue R, Moravick S, Yang JJ. Capacity pooling in hospitals: the hidden consequences of off-service placement. Manage Sci [online ahead of print] 15 Aug 2019; DOI: 10.1287/mnsc.2019.3395.
8 . Kohn R, Harhay MO, Bayes B, Song H, Halpern SD, Kerlin MP, et al. The influence of geographic dispersion on outcomes of hospiitalized pulmonary service patients [abstract]. Am J Respir Crit Care Med 2019;199:A5575.
9 . Lyon SM, Auriemma CL, Cooney E, Kent S, Barg FK, Halpern SD. Defining patient- and surrogate-centered outcomes for critical care research. Am J Respir Crit Care Med 2014;189:A2181.
10 . Centers for Medicare & Medicaid Services Prospective Payment Systems. General information. Baltimore, MD: CMS; 2019 [updated 2019 Mar 8; accessed 2019 Nov 13]. Available from:
11 . Vizient I. 2016 risk adjustment methodology. 2016.
12 . Fine JP, Gray RJ. A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc 1999;94:496509.
13 . Linden A, Mathur MB, VanderWeele TJ. EVALUE: Stata module for conducting sensitivity analyses for unmeasured confounding in observational studies. Statistical Software Components S458592. Boston: Boston College Department of Economics; 2019 [revised 2019 Sep 25; accessed 2019 Nov 13]. Available from:
14 . VanderWeele TJ, Ding P. Sensitivity analysis in observational research: introducing the E-value. Ann Intern Med 2017;167:268274.
*Corresponding author (e-mail: ).

M.O.H. is Statistical Editor of AnnalsATS. His participation complies with American Thoracic Society requirements for recusal from review and decisions for authored works.

Supported by U.S. National Institutes of Health (NIH)/National Heart, Lung, and Blood Institute (NHLBI) grants T32 HL007891, F32 HL139107-01, and K23 HL146894-01, and Agency for Healthcare Research and Quality (AHRQ) K12 HS026372-01 (R.K.), NIH/NHLBI K99 HL141678 (M.O.H.), and Dr. Greysen was supported by NIH/NIA K23 AG045338 (S.R.G.).

This letter has a related editorial.

Author disclosures are available with the text of this letter at

Comments Post a Comment

New User Registration

Not Yet Registered?
Benefits of Registration Include:
 •  A Unique User Profile that will allow you to manage your current subscriptions (including online access)
 •  The ability to create favorites lists down to the article level
 •  The ability to customize email alerts to receive specific notifications about the topics you care most about and special offers
Annals of the American Thoracic Society

Click to see any corrections or updates and to confirm this is the authentic version of record