Healthcare utilization, medical costs and mortality associated with malnutrition in patients with chronic obstructive pulmonary disease: a matched cohort study.

Abstract Objective: Although disease-related malnutrition has prognostic implications for patients with chronic obstructive pulmonary disease (COPD), its health-economic impact and clinical burdens are uncertain. We conducted a population-level study to investigate these questions. Methods: We excerpted data relevant to malnutrition, prolonged mechanical ventilation and medications from claims by 1,197,098 patients which were consistent with COPD and registered by the Taiwan National Health Insurance Administration between 2009 and 2013. These patients were separated into cohorts with or without respiratory failure requiring long-term mechanical ventilation, and each cohort was divided to compare cases who developed malnutrition after their first diagnosis consistent with COPD, versus non-malnourished propensity-score matched controls. Results: The prevalence of malnutrition was 3.8% overall (10,259/287,000 non-ventilator-dependent; 1198/15,829 ventilator-dependent). Propensity-score matched non-ventilator-dependent patients who became malnourished (N = 10,242) had comparatively more hospitalizations, emergency room and outpatient visits, longer hospitalization (all p < .01), and higher mortality (HR = 2.26, 95% CI 2.18–2.34) than non-malnourished controls (N = 40,968). Malnourished ventilator-dependent patients (N = 1197) had higher rates of hospitalization, emergency room and outpatient visits, but shorter hospitalization (all p < .001) and lower mortality (HR = 0.85, 95% CI 0.80–0.93) than matched non-malnourished controls (N = 4788). Total medical expenditure on malnourished non-ventilator-dependent COPD patients was 75% higher than controls (p < .001), whereas malnourished ventilator-dependent patients had total costs 7% lower than controls (p < .001). Conclusions: Malnourishment among COPD patients who were not dependent on mechanical ventilation was associated with greater healthcare resource utilization and higher aggregate medical costs.


Introduction
Malnutrition is "a state resulting from lack of intake or uptake of nutrition that leads to altered body composition and body cell mass leading to diminished physical and mental function and impaired clinical outcome from disease" 1 . There is a growing body of research on the burden that malnutrition imposes on patients and healthcare systems 2,3 . The detrimental impacts of disease-related malnutrition are compounded by its high prevalence and association with suboptimal functional recovery and poor survival 4 . Malnourished people are more likely than are non-malnourished counterparts to be hospitalized, have complications and to die 5 . Malnutrition also imposes considerable financial costs, accounting for between 2% and 10% of annual national healthcare expenditure in European countries 6,7 and the United States 3,8 . In general, malnourished patients receiving medical or surgical treatment had 55% higher medical costs than those without malnutrition, besides having longer hospital stays and needing more care 9,10 .
Chronic obstructive pulmonary disease (COPD) is increasingly recognized as an important problem, due to its heavy healthcare burden worldwide 11,12 ; moreover, becoming malnourished might worsen COPD. In a United Kingdom survey of outpatients with COPD, up to 21% were at risk for malnutrition, most of whom were likely to need nutritional treatment, and the overall prevalence of malnutrition increased with COPD severity 13 ; the outpatients at risk of malnutrition had more hospital admissions and were more likely to die within 6 months 14 . Others have reported that COPD is commonly accompanied by nutrition impact symptoms, which are more severe in nutrition-depleted individuals 15 . Although undernourishment does not contribute to diaphragm weakness in patients with COPD 16 , they have protein metabolism abnormalities that contribute to skeletal muscle alteration 17 . Severe COPD and advancing age can independently contribute to malnutrition 18 . Poor lung function and impaired exercise tolerance were associated with a high prevalence of malnutrition among hospitalized COPD patients, while nutritional status in patients with a COPD exacerbation was related to the duration of hospitalization and the readmission rate 19 . In another study, hospitalized patients with COPD who were underweight or lost weight during follow-up had a higher risk of new exacerbations 20 . Given that COPD patients are known to have systemic comorbidities such as deconditioning, exercise intolerance, skeletal muscle dysfunction, osteoporosis, metabolic impact, anxiety and depression, and cardiovascular disease, with consequently high mortality, it is important to recognize malnutrition early so that timely intervention can be instituted 21 .
Although malnutrition is associated with inimical outcomes in COPD, the magnitude of its overall impact is unknown; most reports on the health-economic burden of malnutrition have been based on data from single institutes or regional registries, with cohorts involving predominantly hospitalized patients 22,23 . Public sector healthcare in Taiwan is reimbursed by universal National Health Insurance (NHI), which covers outpatient, inpatient and emergency care. Patients with acute respiratory failure are initially admitted to an intensive care unit (ICU) and treated with mechanical ventilators. NHI policy is to transfer patients with respiratory failure who have used mechanical ventilation for more than 21 days to a step-down respiratory care center and onward to a general respiratory care ward (unless there is specific indication for remaining in the ICU or subsequent readmission) 24 . Patients receiving mechanical ventilation who have chronic respiratory failure can be discharged to home ventilator use, which NHI also covers. To investigate the importance of malnutrition in patients with COPD at the population level, we used the well validated Taiwan NHI claims database to quantify the impact of malnutrition on healthcare utilization and medical costs.

Study design, setting and purpose
This study analyzed data excerpted retrospectively from the Taiwan NHI Research Database (NHIRD), to investigate the impact of malnutrition on clinical and health-economic outcomes in patients with COPD.  Version -Clinical Modification (ICD-9-CM) diagnostic codes. COPD was defined as insurance claims documentation of three consecutive outpatient visits or a single hospitalization with an ICD-9-CM code consistent with COPD (490.X and 496; Supplementary Table S1) 18 ; the date participants were first given such an ICD-9-CM code was defined as the index date. Potentially eligible patients were subdivided into those with a catastrophic illness registration denoting respiratory failure requiring prolonged mechanical ventilation (ICD-9-CM 518.85; Supplementary Table S1) 24 and those without respiratory failure (most of whom were not dependent on mechanical ventilation). This classification was based on the Taiwan NHI policy that patients who are dependent on mechanical ventilation (i.e. invasive or non-invasive mechanical ventilation for !21 days) can apply for a catastrophic illness certificate that exempts them from co-payments for outpatient or inpatient care 24 . We reasoned that the care settings of patients with versus without ventilator dependence would probably differ, resulting in different treatment costs for either subgroup of patients. Patients who did not fulfill the eligibility criteria were excluded, as were those with an index date during 2009, to ensure that their clinical features and comorbidities had existed for at least 1 year preceding the index date. History of malnutrition was based on ICD-9-CM codes during 1 year preceding the index date that included: 260-262, 263.0, 263.2, 263.8, 263.9, 265.X, 266.X and 268.X, which have been shown to adequately cover the clinical manifestations of malnutrition 25 , and 799.X, based on experience that clinicians might also use these codes to designate malnutrition in certain circumstances, such as cachexia.

Participant selection and outcome variables
Patients with or without ventilator dependence who were diagnosed with malnutrition after the index date (based on the aforementioned ICD-9-CM codes) constituted cases. We calculated the Charlson Comorbidity Index (CCI) scores based on comorbidity profiles during the year before the index date 26,27 . Logistic regression was used to model the risk of mortality in patients with or without ventilator dependence as a function of demographic characteristics (age, sex, residential location), comorbidity profile and CCI scores. Based on the regression results, a 1:4 propensity-score matched control group for cases was identified using an established procedure 28 , with a balanced distribution of covariates among patients with/without malnutrition in the groups with/without ventilator dependence, based on the nearestneighbor approach.
We excerpted demographic information, diagnosis records and NHI claims data relevant to healthcare utilization and clinical outcomes for both malnourished cases and non-malnourished controls from either analytic cohort with or without ventilator dependence, from the index date until death or 31 December 2013, whichever occurred first. The primary outcome was healthcare resource utilization, including annual numbers of outpatient and emergency room visits, frequency and duration of hospitalization, and claim-based aggregate costs associated with medical care from the index date until the end of follow-up, converted from New Taiwan Dollars at the time of analysis to 2013 United States dollars 29 . We also estimated survival at 4-year follow-up for propensity-score matched malnourished cases and non-malnourished controls in the ventilator-dependent and non-ventilator dependent cohorts.

Statistics
Statistical analyses were performed using R-2.11.1 for Windows (R Foundation for Statistical Computing, Vienna, Austria) 30 , run on Microsoft Windows 7 Operating System. We compared demographic features of patients with versus without malnutrition before re-grouping by propensity-score matching. For clinical features including demographic and comorbidity data, and all outcomes, continuous variables were expressed as mean and standard deviation (SD), compared using Student's t-test. Categorical variables were expressed as percentages, and compared using the chisquare test. After propensity-score matching, we analyzed healthcare utilization and costs by applying a two-part model process. Logistic regression was applied to predict the probability of healthcare service utilization, whereas a generalized linear model was used to predict the frequency of healthcare utilization and costs among those with visits or hospital admissions. Covariates entered into the models to predict the healthcare utilization included age, sex and residential area urbanization level. We then used t-tests to examine differences in the regression-adjusted frequency of healthcare utilization and costs between cases and controls. We used Kaplan-Meier analysis to plot survival curves based on the absence or presence of malnutrition in COPD patients with versus without ventilator dependence and compared the results with a log-rank test. Cox proportional hazard regression analysis, incorporating socio-demographic variables, CCI and malnutrition, was applied to assess the relationship between malnutrition and mortality in COPD patients with versus without ventilator dependence.

Ethical compliance
National Taiwan University Hospital Research Ethics Committee approved the study protocol (#201503028W) and waived the requirement for informed consent, based on the anonymity of NHIRD records.

Results
Between 2009 and 2013 the Taiwan NHIRD recorded insurance claims made by 1,197,098 patients who had ICD-9-CM codes compatible with COPD; Figure 1 shows the patient selection flowchart. The analytic cohort without ventilator dependence, selected by applying the inclusion/exclusion criteria from among 1,123,790 patients prescribed medications for COPD, comprised 287,000 patients, of whom 10,259 (3.6%) had malnutrition. From another 73,308 patients with ICD-9-CM codes compatible with both COPD and respiratory failure, 57,479 were excluded and a second analytic cohort included 15,829 ventilator-dependent patients, 1198 (7.6%) of whom had malnutrition. In total, 11,457 of 302,829 patients in both analytic cohorts had malnutrition, a prevalence rate of 3.8%. Table 1 summarizes the demographic and baseline clinical characteristics of the selected patient cohorts with or without ventilator dependence before propensity-score matching. In the non-ventilatordependent cohort, patients with malnutrition were older, predominantly male and proportionally more had CCI > 2 (all p < .001) compared to non-malnourished ones. In contrast, malnourished patients in the ventilator-dependent cohort had sex ratios similar to those without malnutrition, but were significantly younger (p ¼ .005), and a smaller proportion had CCI > 2 (p ¼ .010).
After 1:4 propensity-score matching, the non-ventilatordependent cohort comprised 10,242 malnutrition cases and 40,968 non-malnourished controls, while the ventilatordependent cohort comprised 1197 cases with malnutrition and 4788 controls (Figure 1). Malnutrition cases and nonmalnourished controls had generally similar socio-demographic and comorbidity profiles (Table 2); however, there were some statistically significant differences. In the matched non-ventilator-dependent cohort, COPD patients with malnutrition had a higher prevalence of congestive heart failure (p ¼ .028) than controls, and significantly higher rates of cerebral vascular disease, dementia, peptic ulcer, diabetes and cancer (all p < .001). Malnourished cases in the matched ventilator-dependent COPD cohort had proportionally more diabetes with end-organ damage compared to controls (p ¼ .003), but similar proportions of other comorbidities. Table 3 and Figures 2 and 3 show the 4-year outcomes in the ventilator-dependent and non-ventilator-dependent propensity-score matched cohorts. Compared to non-ventilator-dependent controls, malnourished patients had more hospitalizations, emergency room and outpatient visits, and days hospitalized, amounting to 9.7% of total patient-years hospitalized during the observation period, versus 6.7% for controls (all p < .001) (Figure 2, Table 3). Malnourished patients in the matched ventilator-dependent COPD cohort likewise had more hospitalizations, emergency room and outpatient visits than controls; however, they were hospitalized for fewer days on average and 46.8% versus 52.1% of total patient-years (all p < .001) (Figure 2, Table 3). Malnourished patients in the matched non-ventilatordependent cohort had higher rates of total mortality and inhospital mortality than non-malnourished controls, and almost double the total medical costs (all p < .001), whereas  Abbreviations. COPD, Chronic obstructive pulmonary disease; SD, Standard deviation. a Based on Taiwan National Health Insurance hospitalization claims: "Death in catastrophic illness", "Insurance data termination with no medical records", or discharge record of death or being terminally ill and critically discharged. b Based on Taiwan National Health Insurance hospitalization claims discharge record as death or being terminally ill and critically discharged. malnourished patients in the matched ventilator-dependent cohort had lower rates of total and in-hospital mortality, and slightly lower costs than controls (all p < .001) (Figure 3, Table 3).
Cox proportional hazard analysis showed that malnutrition independently increased the risk for mortality in propensityscore matched non-ventilator-dependent patients (p < .001), whereas malnutrition in the ventilator-dependent cohort was associated with a lower risk for mortality (p < .001) ( Table 4). Both cohorts showed an increasing likelihood of death with advancing age, with men more likely than women to die.

Discussion
Among this nationally representative patient sample, the prevalence of malnutrition after the index date of receiving an ICD-9-CM diagnostic code consistent with COPD was 3.8%. Malnutrition increased healthcare resource utilization and costs, as well as mortality, among COPD patients who were not ventilator dependent. Incongruently, malnutrition in patients with COPD who required prolonged mechanical ventilation was associated with slightly lower overall medical costs and better prognosis. This study, based on national insurance claims data, corroborates findings from smaller institutional cohorts and consequently provides valuable insights about the prevalence and impact of malnutrition in COPD patients at the population level. Our discovery of substantial prevalence of malnutrition comorbid with COPD warrants large-scale, prospective longitudinal population studies based on objectively diagnosed malnutrition.
Congruence with the results of previous studies not only affirms the feasibility of this methodology but also highlights the importance of diagnosing malnutrition in clinical practice. Professional societies recommend screening using validated tools, such as the Mini-Nutritional Assessment, Subjective Global Assessment, Nutritional Risk Screening and suchlike, to identify individuals at risk for malnutrition, followed by definitive diagnosis based on body mass index and body composition metrics 31,32 . However, methodological shortcomings include inadequate statistical power, selection bias towards those amenable to receiving assessment and inconsistent ascertainment methods. An alternative is to identify malnourished patients through medical records of administrative codes relating to malnutrition, which mitigates heterogeneity across institutes, patient populations and healthcare settings 25 . Researchers have used ICD-9-CM codes to estimate the national prevalence and risk factors for malnutrition 4 and healthcare resource utilization in hospitalized patients 33 , affirming the feasibility of this approach. In our study, using ICD-9-CM codes facilitated a more comprehensive analysis of healthcare expenditure in patients with COPD, highlighting that this issue warrants heightened attention and indicating that a population-based approach provides a realistic and balanced view of malnutrition.
Our results suggest not only higher utilization of healthcare resources in malnourished versus non-malnourished patients with COPD, but also imply that this correlation is stronger among those without prolonged ventilator dependence. Disease-related malnutrition has been estimated to cost 15 billion dollars in the United States, with malnourished patients with dementia and depression accounting for 10 billion 2 . The evident association between malnutrition and healthcare utilization by patients with COPD in our study, especially among non-ventilator-dependent patients, was commensurate with previous studies in other patient populations 34 . On the other hand, our finding that malnourished patients who were ventilator dependent had slightly lower medical costs and mortality than adequately nourished counterparts suggests a less straightforward relationship between nutritional status and health outcomes in such patients. One explanation is that the high NHI cost of reimbursing long-term mechanical ventilation might overshadow that imposed by malnutrition. Collectively, these populationbased findings warrant clinical trials of early interventions to prevent malnourishment and improve nutrition in COPD patients.
More than double the risk for mortality in non-ventilatordependent COPD patients with malnutrition highlights the importance of research to understand better how malnutrition interacts with COPD in this setting. Malnutrition in COPD patients with chronic respiratory failure is strongly related to hyperinflation, and non-invasive ventilator treatment in malnourished patients resulted in significant weight gain 35 . In another study, a high-fat, low-carbohydrate oral supplement improved pulmonary function in patients with COPD significantly compared with a conventional highcarbohydrate diet 36 . Malnutrition is also known to play an important role in how elderly individuals with COPD perceive their symptoms 37 . Elderly patients with COPD might complain of lacking energy, which is one of eight items assessed by commonly used tools in the outpatient setting, such as the COPD Assessment Tool 38 ; however, the utility of this metric for additional diagnosis of malnutrition remains unclear. As current clinical guidelines for COPD make few recommendations regarding nutrition 39 , we advocate further research to investigate how malnutrition contributes to increased risk of mortality in patients with COPD, and trials to evaluate the effects of systematic screening and early intervention to prevent malnutrition in patients with COPD.
Our study had noteworthy limitations. Foremost, we did not classify COPD severity based on lung function (e.g. forced expiratory volume in 1 s) as recommended by clinical guidelines for the diagnosis and management of COPD 39 . The Taiwan NHIRD also lacks data on lung function or grading; therefore, we grouped patients according to documented prolonged ventilator dependence, which implied more inpatient care with additional utilization of resources (Figures 2 and 3). Second, selecting malnutrition cases based on administrative data relies on physicians' diagnosis of malnutrition, rather than the objective examination of patients' current physiologic and nutritional states; therefore, malnutrition may be under-diagnosed, and assessing its impact more accurately might require further research 40 . Nevertheless, malnutrition prevalence of 3.8% in patients with COPD is commensurate with other reports. Furthermore, this methodology would allow comparisons across different patient populations and countries based on the same ICD-9-CM coding. Third, mortality was documented indirectly, based on NHI hospitalization claims data; nevertheless, since Taiwan NHI covers almost all healthcare expenses, most deaths among included patients would be counted. Fourth, malnutrition severity was not assessed, which precluded quantitative estimation of the differential impact of increasingly severe malnutrition on healthcare utilization and burden. Therefore, although this study highlights the significant impact of malnutrition in COPD patients, further research is required to ascertain the magnitude of its impact. Last, we were unable to investigate the role of nutritional supplementation because the NHIRD does not record this information. The main potential biases included the probability of comorbidities that might also entail resource use and associated care costs because patients with COPD have a high probability of comorbidities during their disease course. Although the costs in a substantial proportion of patients in this study group might not be attributed to COPD alone, the propensity-score matching design eliminated this problem in comparisons between patients with and without malnutrition.

Conclusions
This population-based study of a nationally representative cohort of patients with COPD affirms that non-ventilatordependent patients with malnutrition use more healthcare resources, impose higher healthcare costs and have worse survival than those who are not malnourished. Our findings highlight the priority of diagnosing malnutrition in patients with COPD early, and of implementing prompt and timely management to diminish its associated healthcare burdens.

Declaration of funding
Abbott Nutrition, Taipei, Taiwan provided funding to support manuscript editing and publication. have disclosed that they are employees of Abbott Nutrition. All other authors report no potentially conflicting interests. CMRO peer reviewers on this manuscript have no relevant financial or other relationships to disclose.