Risk factors and outcomes associated with chronic obstructive pulmonary disease exacerbations requiring hospitalization

1Centre for Clinical Epidemiology and Evaluation, Vancouver Coastal Health Research Institute and Respiratory Medicine, Vancouver General Hospital; 2Division of Respiratory Medicine, University of British Columbia; 3Division of Respiratory Medicine, St Paul’s Hospital, Vancouver, British Columbia Correspondence: Dr John Swiston, Division of Respirology, University of British Columbia, Vancouver General Hospital, 2775 Laurel Street, 7th Floor, Vancouver, British Columbia V5Z 1M9. Telephone 604-875-4122, fax 604-875-4695, e-mail swiston@interchange.ubc.ca Chronic obstructive pulmonary disease (COPD) is a major cause of disability and death in Canada, and throughout the world. Approximately 500,000 Canadians older than 35 years of age have been diagnosed with COPD, and it is estimated that an almost equal number of middle-aged Canadians may also have COPD, but are not yet aware of it (1). Acute exacerbations of COPD (AECOPD) are the most frequent cause of medical visits, hospitalization and death for COPD patients (2). While the length of stay (LOS) for hospitalizations as a result of AECOPD has been decreasing in Canada in recent years (3,4), the total number of hospitalizations has increased (5). Despite the tremendous burden of hospital admissions, little is known regarding the factors associated with hospital admission and readmissions. In a recent systematic review (6), we examined the published literature pertaining to risk factors for hospitalization and readmission of COPD patients from 1966 to 2006. A review of 17 relevant studies evaluating the hospitalization of oRiginal aRticle

Can Respir J Vol 16 No 4 July/August 2009 e44 COPD patients showed significant differences in risk factors reported and associations identified, and found few risk factors that have consistently demonstrated a clear association with admission and readmission.As an example, long-term home use of supplemental oxygen (LTOT) and inhaled corticosteroids were independently associated with admission or a shorter time to first readmission for AECOPD in some of the reviewed studies (7)(8)(9)(10)(11).However, in other studies, these associations were not significant or they did not remain significant after an adjustment for potentially confounding factors (7,12,13).Similarly, patient management by a respirologist, lower forced expiratory volume in 1 s (FEV 1 ) and oral corticosteroid use were also associated with a higher risk of readmission in some studies (7)(8)(9)(13)(14)(15)(16)(17) but not others (7,9,14,18,19).These inconsistencies and a need for better definition of the determinants of health in COPD patients within the context of the Canadian health care system, led us to undertake the current study.
The aims of the present study were to ascertain the rates of hospital readmissions for AECOPD among a cohort of patients previously hospitalized with the same diagnosis and to identify risk factors associated with recurrent hospital admissions for AECOPD.

Design and subjects
The charts of all patients who were admitted for an AECOPD to three general hospitals in the Vancouver, British Columbia, area over a 20-month period between April 1, 2001, and December 31, 2002, were retrospectively reviewed.The index hospitalization was defined as the patient's first admission to an acute care hospital with a primary diagnosis of an AECOPD using the International Classification of Diseases 9th Revision -Clinical Modification (ICD-9-CM) coding.The following ICD-9-CM codes were included in the present study: 490, 491 (491.0,491. 1, 491.2, 491.20, 491.21), 492 and 496.The index discharge date was defined as the date on which the patient was discharged for the AECOPD.Both COPD patients with a first-episode exacerbation and repeat admissions for subsequent exacerbations were included.
Patients were excluded from the study if they were older than 95 years of age or had a history of asthma.Admissions were excluded from analysis if the hospitalization was primarily for elective surgery and the diagnosis of COPD was coincidental, if the admission was the result of a transfer from another acute care hospital, if the patient was discharged to another acute care hospital or if the patient signed themselves out against medical advice.The protocol was approved by the Ethics Committees of the participating hospitals as well as the University of British Columbia's Clinical Ethics Committee (Vancouver, British Columbia).

Characteristics of the three general hospitals
Hospital A is the province's largest hospital, with 955 acute care beds.Hospital B has 440 acute care beds, and hospital C has 140 acute care and 100 extended care beds.All three hospitals offer primary and secondary care, are affiliated with the University of British Columbia (Vancouver, British Columbia) and accept referrals from throughout the province, with primary catchments being within the greater Vancouver area.Hospitals A and B are also tertiary care facilities.All three hospitals offer access to specialist services including internal medicine and respirology.The ICD-9-CM diagnostic codes used to classify the primary admission diagnosis were the same for all three hospitals.

Data collection
Data were retrospectively collected from health records by a contributing author (KB) using a standardized data collection instrument developed for the project.Information regarding the index admission was obtained from progress notes made by physicians, as well as charting records and notes from nurses and other allied health care workers.Information bias was limited by the use of a set of 'practice' medical records, formal inclusion criteria, the use of a standardized data abstraction form and reviewing coding rules with health experts (20).
Variables were selected for analysis based on previous studies pertaining to readmission for an AECOPD and other chronic diseases (8,13,16,17,21).Analyzed variables included demographics (eg, age and sex), admission statistics (number of admissions during the study period, dates of index admission and discharge, LOS during the index admission and date of first readmission after discharge from index admission), social characteristics (smoking status, occupation, living situation, home care status), Charlson comorbidity indices within one year preadmission (22), evidence of previous lung infections within one year of index admission, preadmission COPD therapy (eg, medications, home oxygen therapy, inhaled or systemic corticosteroid); functional status (mobility, activities of daily living); spirometry; and discharge care (referrals to physiotherapy and occupational therapy, as well as follow-up arrangements if any, to a family physician, an internist and/or a respirologist).Charts were also reviewed for documentation of previous participation in a COPD education program (including individual teaching, small-group sessions, video and audio tapes, and problem-solving sessions).
Comorbidity was quantified according to the Charlson index (22), which contains 19 categories of comorbidity that are primarily defined using the ICD-9-CM diagnosis codes.Each category has an associated weight that is based on the adjusted risk of one-year mortality.The overall comorbidity score reflects the cumulative increased likelihood of one-year mortality -the higher the score, the more severe the burden of comorbidity.

Statistical analysis
Sample size estimations suggested that 113 patients would be sufficient in the multiple logistic regression analysis to demonstrate an OR of 2.0 (between individuals readmitted and those not readmitted) -assuming a correlation of 0.2 between the covariates -at a significance level of 5% with 95% power (23).Values for continuous variables were reported as mean ± SD unless otherwise stated.Because the primary end point was readmission, data from the 34 patients who died during the study period (11%) were excluded from the analysis.
The association between variables and the risk of readmission were initially assessed with univariate logistic regression.Variables significant at P≤0.10 were further combined and analyzed by stepwise logistic regression.Independent variables with high percentages of missing values (more than 10%) were not included in the analysis.Hospitals were considered as potential covariates and were forced into the regression equation as dummy variables before testing other variables and could not be removed thereafter.Variables that retained significance were then entered into a stepwise multivariate regression analysis to compute the final predictive model for readmission outcome.All statistical analyses were performed using SPSS version 12.0 (SPSS Inc, USA) for Windows software (Microsoft, USA).

Patient characteristics and readmission risk
Between April 1, 2001, and December 31, 2002, 310 patients admitted to the three hospitals with an AECOPD met the study inclusion criteria (Table 1).The baseline characteristics of the cohort are shown in Table 2.One hundred sixty-six of the patients (54%) were men, with a mean age of 74±12 years.The cohort had a mean FEV 1 % predicted of 43±16.Thirtyeight per cent of the subjects were current smokers, with an exposure ranging from 23 pack-years to 172 pack-years.Thirtyone per cent of patients were using home oxygen (O 2 ) before admission, 56% had other chronic respiratory disease, 31% had a history of lung infections (defined operationally based on the history documented in the chart as having any lung infection such as pneumonia or Mycobacterium tuberculosis within one year before the index admission), 37% had a history of congestive heart failure (CHF) and 14% had diabetes.
Of the 310 patients admitted to hospital for an AECOPD, 119 (38%) had at least one subsequent hospital readmission for an AECOPD, 29 of which had two (9%), and 23 (7%) had three or more hospital readmissions -all verified from medical records.Among patients who were readmitted to hospital during the study period, the mean duration from the index admission to the first readmission was 5±4.08 months (n=80).The mean duration from second to third readmission was shorter (2.57±2.65 months [n=26]) (Table1).
The medications prescribed before, during and following hospitalization are shown in Table 3. Care at the time of admission (first 24 h) was analyzed using the total number of admissions (n=503).The present analysis, therefore, included some patients more than once (ie, individuals who were readmitted); however, the purpose of the present analysis was to evaluate health care delivery -not the features of individual patients.Patient characteristics preadmission and at discharge were evaluated using the number of patients (n=310) -not the number of admissions.Twenty-three per cent of patients were taking antibiotics before admission, suggesting that almost one-quarter of patients were failing an attempt at outpatient therapy for AECOPD.Twenty-one per cent of patients were using oral corticosteroids before admission.Inhaled corticosteroids were used by 45% of patients before admission, while 39% of patients were prescribed inhaled corticosteroids at discharge.Oral or parenteral corticosteroids were administered in the first 24 h in only 59% of AECOPD admissions.No association was found between the use of oral corticosteroids preadmission and the risk of readmission(s).
Univariate analysis was performed with baseline characteristics as independent variables and hospital readmissions as the dependent variables.Significant risk factors for readmission to hospital included home O 2 use preadmission (OR 2.62; 95% CI 1.58 to 4.32; P=0.000), a history of lung infection within one year before the index admission (OR 2.06; 95% CI 1. 25  Variables with a P>0.05 but P<0.1 (body mass index, LOS and having home care nursing) were considered for the multiple There was no significant interaction between variables in the logistic regression analysis.

Comparison of the three hospitals surveyed
Comparative analysis among the three hospitals identified a significant difference in readmission rates.Hospital A had the highest rate of readmission (54%), while hospital C had the lowest (18%) and hospital B was in between (37%).To explore underlying reasons for the differences in outcomes between the hospitals, both patient and health care-related factors were evaluated (Table 6).Patient characteristics compared among the three hospitals included demographics, premorbid health characteristics and social characteristics.Fewer men were admitted to hospital C (39%) compared with 59% and 58% for hospitals A and B, respectively.Hospital C also had the highest percentage of patients admitted who were not Canadian born (55%) compared with hospitals A (29%) and B (17%).Regarding premorbid health characteristics, the average FEV 1 % predicted was lowest in hospital A (33±10) and highest in hospital C (56±19), suggesting an inverse relationship between FEV 1 and the rate of hospital readmission in the three hospitals.There was also a higher percentage of home O 2 use preadmission among patients admitted to hospital A (40%) compared with hospital B (27%) and hospital C (22%).Furthermore, CHF was seen more often in hospital A (48%) compared with hospitals B (29%) and C (31%).Of the social characteristics evaluated, no trends that correlated with the differences in the rates of readmission among the three hospitals were identified.Health care and delivery characteristics that were evaluated to explore the differences in outcomes among the three hospitals included admitting services, respiratory medications prescribed in the first 24 h of admission and follow-up care arrangements made at the time of discharge.Although all three hospitals primarily admitted AECOPD patients to respirology and internal medicine departments, hospital A had the highest rate of admission to family practice services (29%) compared with both hospital B (3.2%) and hospital C (10.4%).Antibiotic use in the first 24 h of admission was higher in hospital C (88%) than in hospitals A and B (59% and 62%, respectively).Hospital A, with the highest readmission rate, had the lowest rate of systemic corticosteroid (combined oral and parenteral) administration in the first 24 h of admission (45% versus 73% versus 59% for hospitals A, B and C, respectively).Overall, follow-up arrangements made with physicians at the time of discharge did not seem to positively affect the rate of readmission.

DiSCuSSion
The present study described the outcomes of 310 patients who experienced at least one AECOPD resulting in admission to one of three Vancouver area hospitals over a 20-month period during 2001 and 2002.Readmission rates for COPD patients were high.During the study period, 38% of subjects were readmitted at least once and the mean duration from the index admission to the first readmission was only 5±4.08 months.A previously described cohort of 345 COPD patients (13) admitted to hospital for COPD exacerbations and followed for readmission in Spain found a similar rate.The authors reported a first readmission rate of 63% over a mean follow-up period of 1.1 years and a median time to first readmission of 186 days (13).The high frequency of readmission and the short time to readmission found in our study and that of Garcia-Aymerich et al (13), highlight the social and economic burden that COPD exerts on society.

Patient characteristics and readmission risk
Despite evaluating a wide range of both modifiable and nonmodifiable potential risk factors for hospital readmission, only a few clinical factors were found to be independently associated with COPD readmission in the present study.The preadmission use of home O 2 was found to be associated with a higher risk of readmission in both univariate and logistic regression analyses.A number of previous studies (7,9,11) have also reported home O 2 as a risk factor associated with an increased risk of admission or frequent readmissions.It is likely that home O 2 use is a marker of disease severity, and individuals with more advanced disease are at higher risk for readmission.Previous lung infections (within one year before index admission) and other chronic respiratory disease were also both associated with an increased risk of readmission in the logistic regression analyses.The presence of other chronic respiratory disease may increase susceptibility to readmission simply by compounding the impairment in lung function, or may impact overall systemic fitness and, therefore, the global ability to cope with acute illness.Individuals with documented previous lung infections within one year before admission may be at higher risk of readmission because increased frequency of chest infections leads to more frequent COPD exacerbations.With increased COPD severity, the likelihood of admission during exacerbations increases.
Patients readmitted to hospital were more likely to have shorter LOSs.However, previous studies (21,24) have indicated that increased disease severity is associated with prolonged LOS.It may be that in our cohort, shorter LOS was an indication of premature discharge before the clinical and social aspects of the admission were fully resolved and, thus, patients were more likely to return to hospital.It is possible that very short hospital admissions do not allow for appropriate convalescence and discharge planning, while very long LOSs represent individuals with severe disease requiring long-term therapy or those with more precarious social circumstances that make discharge planning more difficult.
No association was found between current smoking and the risk of COPD admission or readmission.This finding supports the result of a previous study by Kessler et al (9) that found no differences between current smokers and the risk of COPD readmission.An explanation has been proposed by Anthonisen (23), who suggested that severely ill patients with COPD spontaneously quit smoking probably in response to their symptoms and disability.Prospective studies including a wider range of COPD severity would help to clarify this issue.
The FEV 1 % predicted was not a predictor of hospital readmission for an AECOPD in the present study.Several studies have similarly reported a lack of association between FEV 1 % precicted and COPD exacerbations and admissions (9,25,26); however, others (13,15,27,28) have found lower FEV 1 or severity of FEV 1 impairment to be predictive of a higher risk of COPD admission and readmission.It is possible that the lack of association in the present and other studies was the result of inadequate statistical power rather than a true lack of association.Alternatively, the lack of association between FEV 1 and hospital readmission may be due to relatively limited variation in the FEV 1 across the study population.

Comparison of the three hospitals
The rates of hospital readmission varied among the three hospitals.Comparison of the patient and health care characteristics provided an opportunity to explore possible explanations for the differences in readmission rates.The cohort of patients admitted to the hospital with the highest rate of readmission (hospital A) appeared to have more severe COPD as demonstrated by the lowest mean FEV 1 and highest percentage of home O 2 use before admission.It may be that severity of COPD is the primary explanation for the variation in the rates of readmission.In support of this conclusion, a recent study published by Gonzalez et al (29) in 2008, found that patients with LTOT had a lower FEV 1 % predicted, and the use of LTOT was independently related to the risk of readmission for patients with moderate to severe COPD.
Overall, the hospital with the highest rate of readmission for AECOPD (hospital A) had the lowest rate of systemic corticosteroid administration in the first 24 h of admission.This variation in the standard of care may, in part, explain the difference in hospital readmission rates.If this is the case, this finding would support the implementation of standardized care pathways for AECOPD based on current best management guidelines.It is interesting to note that hospital A also had the highest percentage of AECOPD admissions to the family practice service compared with other hospitals.Whether the relationship between readmission rates and admitting service is a consequence of immediate corticosteroid use or a result of more general overall care differences remains unclear and requires a more direct evaluation.

Limitations
The primary limitation of the present analysis is the retrospective design.As with all retrospective studies, caution must be exercised when interpreting the results because selection, information and recall bias cannot be completely accounted for.It is possible that not all readmissions were captured because some patients may have been readmitted to a hospital other than the three surveyed in the present study.However, the three hospitals studied provide all of the acute care for the Vancouver area; thus, it is unlikely that a significant number of readmissions to other hospitals were missed.The frequency of exacerbations in the year before admission was likely underestimated because only exacerbations for which a discharge report or a hospital record was available were included in an attempt to limit recall bias.Finally, the diagnosis of AECOPD relied on clinical assessments that were not standardized across time and location before the chart review.

ConCLuSion
We found a high rate of readmission for AECOPD within a 20-month study period.Furthermore, there was significant variability in the readmission rate among three hospitals within the same city.This variability may be a consequence of differences in the populations served by each or may reflect differences in health care delivery that impact patient outcomes.Prospective studies are warranted to explore these possibilities further because the former may identify modifiable patient characteristics while the latter may identify modifiable health care characteristics.A history of lung infections within one year before admission, the presence of other chronic respiratory disease, the use of home O 2 before admission and longer duration of admission were significantly associated with a higher likelihood of being readmitted to hospital.These characteristics may be markers of COPD severity or identify individuals with overall more precarious health and/or social circumstances.These results raise the possibility that these variables could be used as identifiers of patients at highest risk for frequent exacerbations and readmissions.DiSCLoSuRe: The present study was funded by an unrestricted educational grant from GlaxoSmithKline.

aCKnoWLeDGeMenTS:
We are grateful to Dr Penny Brasher, Dr Michael Schulzer and Dr Mohsen Sadatsafavi for their data management and statistical advice.Dr FitzGerald is a MSFHR Distinguished Scholar Recipient as well as a BC Lung/CIHR Scientist.

TaBle 4 Comparison between patients experiencing one admission versus one or more readmissions (univariate analysis)
*Indicates P values less than 0.1 (cutoff for inclusion in the multiple regression analysis).† Indicates variables with greater than 10% missing data.COPD Chronic obstructive pulmonary disease; FEV 1 Forced expiratory volume in 1 s; O 2 Oxygen; PT/OT Physiotherapy/occupational therapy

TaBle 5 Final multivariate models of risk factors for readmission to hospital for acute exacerbations of chronic obstructive pulmonary disease
Can Respir J Vol 16 No 4 July/August 2009 e47