The association of the number of comorbidities and complications with length of stay, hospital mortality and LOS high outlier, based on administrative data
© The Japanese Society for Hygiene 2008
Received: 23 July 2007
Accepted: 3 December 2007
Published: 29 March 2008
With greater concern for efficient resource allocation and profiling of medical care, a case-mix classification was applied for the per-diem payment system in Japan. Many questions remain, one of which is the role of comorbidity and complication (CC) in grouping logic. We examined the association of the number of CC with the length of hospital stay (LOS) and hospital mortality as well as the proportion of LOS high outliers in 19 major diagnostic categories (MDCs).
This study was a secondary data analysis embedded in a government research project, including anonymous claims and clinical data during a 4-month period from July 2002. Every 19 MDC, LOS, hospital mortality or proportion of LOS high outliers was compared by the number of CC and presence of any procedures.
From 82 special function hospitals, 241,268 patients were enrolled in this study. Among all patients, 50.5% were identified without any CCs, 32.4% with one or two, 13.4% with three or four, and 3.7% with over five CCs. The overall mean LOS was 22.15 days and hospital mortality 26.05 cases per 1,000 admissions. In any MDC, LOS and the proportion of outliers increased as the number of CC rose. The mortality rate increased prominently in the respiratory system and the hematology system.
This study demonstrated that the occurrence of more CC caused longer LOS and higher mortality in some major disease categories. Further study will clarify the association of the weighted CC with resource use through controlling procedures specific for MDC.
The role of comorbidities and complications (CC) in inpatient medicine has been an intense area of investigation, due to their impact on resource use, mortality, functional status, quality of life, and delivery of health care in Western societies [1–6]. Research on the impact of CC on healthcare utilization and quality of medical care is needed, as this is important to hospital payment systems and medical outcome studies [7, 8]. Many health-related problems increase with age, especially with respect to the number of chronic conditions . As the proportion of chronic conditions increases across populations, practice behaviors will change that will impact hospital costs. Meanwhile, innovative procedures have been advocated for the elderly [9, 10]. Under these circumstances, the impact of CC or age on hospital resource utilization has been examined for diseases such as diabetes mellitus or for hospitalizations of trauma patients [11, 12].
In response to increasing costs associated with the rapid evolution of healthcare technology, cost containment policies and case-mix classification systems have advanced worldwide over the past 20 years [7, 8]. However, problems have emerged in situations where some patients are under-reimbursed if they have more CCs and have consumed more hospital resources than those with fewer CCs. Munoz et al.  reported major inequities in Diagnosis-Related Group (DRG) prospective payment systems for pediatric patients. Jencks et al.  showed that the number of recorded diagnoses was no higher for patients who died than for those who survived. These studies focused on selected conditions, such as pediatric illnesses, cardiovascular disease, and pneumonia.
In this study, our goal was to demonstrate the association of CCs with resource use and outcome by major diagnostic category (MDC) and treatment group (medical or surgical) using a large administrative database developed for the Japanese case-mix classification system (Diagnosis Procedure Combination; DPC). This type of systematic research has never been performed on this scale, and the results may have some important policy implications, in which MDC CCs may be considered in prior risk adjustment for payment or outcomes, such as resource use or mortality. The aims of this study were to generate descriptive statistics of CCs and to profile the association of length of stay (LOS) and hospital mortality with the number of CCs through the stratification of MDCs and treatments. Furthermore, we identified the relationship between the proportion of LOS high outliers and the number of CCs for each MDC.
Materials and methods
We conducted a secondary data analysis of a government research project on DPC development. Anonymous claims information and clinical data were provided by the Ministry of Health, Labor and Welfare (MHLW) through a research contract. Clinical data and claims information, merged into a standardized electronic format, were gathered by the MHLW for 266,677 patients who were discharged from 82 academic hospitals (80 university hospitals, National Cancer Center, and National Cardiovascular Center) between 1 July 2002 and 31 October 2002. The Japanese MHLW permits up to seven CCs (four comorbidities and three complications) to be listed in the DPC dataset. From this original dataset, we selected cases with a LOS of up to 365 days and excluded patients who died within 24 h of admission. A total of 241,268 patients were enrolled in the study. We categorized patients into three groups according to the number of CCs documented (absent, 1 or 2, 3 or 4, and 5 or more), based on a study by Munoz et al. .
The proportion of comorbidity and complication category by presence of procedure and MDC (%)
Major diagnostic category
One or two
Three or four
Ear, nose, mouth, and throat system
Digestive tract, hepatobiliary and pancreas system
Muscuoskeletal and connective tissue system
Skin and subcutaneous tissue system
Endocrine, nutrition and metabolic system
Kidney, urinary tract and male reproductive system
Pregnancy, childbirth, puerperium and female reproductive system
Injury, burns, poisonings and toxic effect of drugs
Mental health system
Patient characteristics were analyzed in terms of gender, age (under 15, 15–64, and 65 years or older), and number of procedures. The association between CC category and each of these patient characteristics was assessed using Fisher’s exact test. The association between these patient characteristics and MDC was also assessed.
The mean age, LOS, and mortality rate per 1,000 admissions were calculated for each CC category and compared using analysis of variance (ANOVA). The mean LOS (days) and mortality rate (per 1,000 cases) were illustrated using spider-radar charts, stratified by treatment group (surgical or medical) and MDC. ANOVA was used to compare the mean LOS and mortality rate by treatment group for each MDC.
To define the proportion of high LOS outliers, we identified the 95th percentile of LOS for each MDC and categorized patients beyond that LOS into eight groups (e.g., four CC groups by two treatment groups). We then calculated the proportion of high LOS outliers (the numerator is the number of LOS outliers and the denominator is the number of all patients in the eight groups) and demonstrated it with a broken line for each MDC. Statistical analyses were performed using SPSS version 14. All reported P values were two-tailed, and the level of significance was accepted as less than 0.05.
The most frequently documented MDC was the digestive track, hepatobiliary, and pancreatic diseases (45,080 cases, 18.7% of all study cases) and the least frequently documented MDC was systemic infections (1,147 cases, 0.5%). Eye diseases were more common in the surgical group (17,974 cases, 80.7% of all eye diseases), while patients with mental disorders received no surgical treatment. When stratified by CC category, pediatric diseases had the highest proportion of no CCs (64% in the medical group and 70% in the surgical group) and cardiovascular diseases (32% in the medical group and 31.5% in the surgical group) had the lowest proportion of no CCs. On the other hand, hematological diseases (8.1% of medical patients and 12.7% of surgical patients) and systemic infections (7.6 and 22.7%, respectively) had the highest proportion of patients with five or more CCs (Table 1).
The proportion of comorbidity and complication category by age, gender, and presence of surgical procedure (%)
One or two (%)
Three or four (%)
Five or more (%)
Under 15 years
65 years or more
Number of procedures*
Descriptive characteristics of study variables by comorbidity or complication category and proportion of comorbidity and complication category by mean age, LOS, and mortality, stratified by treatment group
Comorbidity and complication category
One or two
Three or four
Five or more
Mean age* (SD)
Mean LOS* (SE)
Mean mortality per 1,000 cases* (SE)
This study describes the characteristics of patients with CC and the association of LOS and mortality rate with the number of CCs by analyzing a large administrative dataset from Japan. To our knowledge, this is the first study to provide a profile of LOS, hospital mortality, and proportion of LOS high outliers across several diseases from a large database. In all MDCs, both the LOS and the proportion of outliers increased as the number of CCs increased, and the increase was particularly prominent for neonatal and pediatric diseases. Mortality was particularly high among the higher CC categories for respiratory diseases, digestive tract, hepatobiliary and pancreatic diseases, and hematological diseases. Therefore, the number of CCs should be taken into consideration in risk adjustment for mortality. However, the mortality rate did not increase with higher CC categories for eye diseases, skin and soft tissue diseases, mental disorders, or in the pediatric surgical group.
There were several limitations to this study. First, we gathered information from patients who were discharged during only a 4-month period in 2002. Claims data, including some clinical variables, are now being collected throughout the year, so that it will soon be possible to produce this type of study with a larger database. Second, coding accuracy and quality were not taken into consideration. At the start of this study, comorbidity was defined by the MHLW as an associated disease or disorder at admission, regardless of whether that condition was acute or chronic and stable; complication was defined as events occurring unexpectedly or owing to a planned procedure after admission. Both of these were listed separately in the dataset. Although there may be no case-mix classification system in the world where the quality of coding can be assured without chart review, coding guidelines and coder training have been promoted by the Japanese Society of Medical Record Administration. The peer review organization system for coding behavior and DPC creep will be in demand, just like in other countries that are already utilizing those kinds of case-mix classification systems [16–19]. Third, there was a limitation of coding slots in the DPC dataset, whereby only seven secondary diagnoses (four comorbidities and three complications) could be listed. However, the dates and the amount of life-support care or pharmaceutical agents, which may serve as proxy data for some comorbidities or complications, were electronically collected in this DPC database . A more detailed and promising analysis of this data is expected in the near future.
Given the paucity of this kind of analysis in the literature, it is useful to document the mean LOS, mortality rates, and proportions of LOS high outliers by MDC and treatment. In general, economic incentives may induce problems like creeping or changing of coding response, whereby institutions may list more CCs or truncate chronic diagnoses unrelated to resource use, outcome, or payment . As a result, the number of recorded diagnoses on the medical chart may not be significantly higher for patients who die than for those who survive, and the number of CCs would not be a reliable index for predicting complications or hospital mortality. Therefore, setting aside the problem of coding accuracy, this study provides basic but instructive suggestions, as the results were free of reporting bias, due to the lack of opportunities for the up-coding of CCs before the start of the DPC-based payment system.
Overall, the mean LOS in Japan may be longer than that in Western countries. Japanese hospitals accommodate patients with both acute and sub-acute or chronic illnesses, functions that are typically performed by different facilities in Western countries . Table 3 shows that the overall mean LOS was 22.15 days (20.39–42.52 days in medical DPC and 24.02–48.30 in surgical DPC), which is two to six times longer than any other country in the OECD Health Data . These hospitals may not be representative of Japanese acute care hospitals, but these results may more accurately reflect mortality and resource consumption, which would not be captured in Western data that do not include data from external care facilities and other auxiliary health facilities. It might be the strength of this study.
In our study, as the number of CCs increased, resource use indices tended to increase for all MDCs, which corresponded to the results of Munoz et al. . They reported that hospital cost, LOS, percentage of LOS outliers, and mortality increased as the number of CCs per patient increased, even for patients categorized into medical non-complications and comorbidity-stratified DRG groups, resulting in financial risk for hospitals without any DRG adjustments based upon CCs. In another study by Munoz et al. , major inequities in the DRG prospective hospital payment system were confirmed for pediatric patients, generating a financial burden for hospital management. For example, if hospital cost correlates positively with LOS, hematological diseases or systemic infections may cause more financial loss, while eye diseases may not. Such comparisons have policy implications, and further studies are needed to examine the presence of cost-profit differences in Japan’s DPC system.
The number of CCs may also be selected as a risk factor for LOS or hospital mortality for respiratory diseases, as well as for digestive track, hepatobiliary, and pancreatic diseases. The number of procedures specific for these diseases, clinical severity, or difficulty of executing a procedure in terms of experience or time consumed may affect LOS, mortality, or LOS outlier among the MDCs. Further evaluation will be needed to describe the association of these clinical variables with the number of CCs or some targeted CCs specific for every MDC.
The first key of DPC classification is principal diagnosis and types of procedures or CCs determined by the DPC group . All of these determinant elements are listed in the DPC definition table where many kinds of CCs are also contained, ranging from chronic stable illnesses, such as diabetes mellitus without organ damage, to acute or critical conditions, such as cardiogenic shock [14, 15]. Each MDC includes several CCs determined by the opinion of experts from the relevant specialties. Our research team was requested to identify the CCs responsible for more resource use or higher mortality, while controlling for variables affecting those indices, such as demographics and treatment. According to the report of resource use variation in cardiovascular diseases and malignant respiratory or intestinal neoplasms, CCs had less incremental effect on the proportion of variance in LOS or total charge than other variables, such as treatment type and intensive or neo-adjuvant therapy . Among these diseases, CCs of gastric or colonic neoplasms explained more variation than those of any others. Some CCs are strongly associated with other CCs, such as hypertension with atherosclerosis. Therefore, further analyses of this kind will be needed to identify CCs that consume more resources, particularly for musculoskeletal diseases and neonatal disease. Through this type of systematic investigation, we can readily answer questions such as “Is the number of CCs correlated with resource use or outcome?” or “Which factors have the greatest impact on LOS or per diem payment: the number of CCs, a specific CC, or certain treatments?” In making decisions about payments, answering these types of questions will facilitate and improve financial allocation. In addition, health policy makers could examine the extent to which the number of CCs can explain variations in resource use or outcomes, enabling a systematic comparison of healthcare performance across MDCs.
In conclusion, we assessed the association of the number of CCs with LOS, hospital mortality, and the proportion of LOS high outliers. In all MDCs, LOS and the proportion of outliers increased as the number of CCs increased. This study demonstrated that the number of CCs should be taken into consideration in risk adjustment for mortality, especially for respiratory diseases, digestive tract, hepatobiliary and pancreatic diseases, and hematological diseases. Mortality rates were not associated with CCs for eye diseases, skin and soft tissue diseases, mental disorders, and in the pediatric surgical group. Further studies are needed to investigate the type of CCs that impact outcomes and resource use, as well as to assess the impact of CCs on treatment selection. Calculating resource use or mortality with weighted CCs, comparative profiling of CCs, and determining associations of CCs with financial burden across MDCs may all play important roles in policy making for an equitable payment system.
This original article is a product of our research team for developing and refining the Japanese case-mix classification (DPC) in cooperation with the Ministry of Health, Labor and Welfare. This research study was led and funded by the Ministry. None of the members have any relevant conflicts of interest.
- Ginsen R, Hoeymans N, Schellevis FG, Ruwaard D, Satariano WA, van den Bos GA. Causes and consequences of comorbidity: a review. J Clin Epidemiol. 2001;55:661–74.Google Scholar
- Evans RL, Hendricks RD, Lawrence KV, Bishop DS. Identifying factors associated with health care use: a hospital-based risk screening index. Soc Sci Med. 1988;27:947–54.PubMedView ArticleGoogle Scholar
- Romano PS, Roos LL, Jollis JG. Adapting a clinical comorbidity index for use with ICD-9-CM administrative data: differing perspective. J Clin Epidemiol. 1993;46:1075–9.PubMedView ArticleGoogle Scholar
- Melfi C, Holleman E, Arthur D, Katz B. Selecting a patient characteristics index for the prediction of medical outcomes using administrative claims data. J Clin Epidemiol. 1995;48:917–26.PubMedView ArticleGoogle Scholar
- Starfield B, Lemke KW, Bernbardt T, Foldes SS, Forrest CB, Weiner JP. Comorbidity: implication for the importance of primary care in ‘Case’ management. Ann Fam Med. 2003;1:8–14.PubMedView ArticleGoogle Scholar
- Zhan C, Miller MR. Administrative data based patient safety research: a critical review. Qual Saf Health Care. 2003;12:58–63.View ArticleGoogle Scholar
- Fetter RB. Casemix classification systems. Aust Health Rev. 1999;22:16–34.PubMedView ArticleGoogle Scholar
- 3M Health Information Systems. All patient diagnosis related groups (AP-DRGs) ver.12.0 definition manual. Wallingford: 3M Health Information Systems; 1994.Google Scholar
- Clark MA, Bakhai A, Lacey MJ, Pelletier EM, Cohen DJ. Clinical and economic outcomes of percutaneous coronary interventions in the elderly. Circulation. 2004;110:259–64.PubMedView ArticleGoogle Scholar
- Harrell AG, Lincourt AE, Novitsky YW, Rosen MJ, Kuwada TS, Kercher KW et al. Advantages of laparoscopic appendectomy in the elderly. Am Surg. 2006;72:474–80.PubMedGoogle Scholar
- Struijs JN, Baan CA, Schellevis FG, Westert GP, van den Bos GA. Comorbidity in patients with diabetes mellitus: impact on medical health care utilization. BMC Health Serv Res. 2006;6:84.PubMedView ArticleGoogle Scholar
- Bergeron E, Lavoie A, Moore L, Clas D, Rossignol M. Comorbidity and age are both independent predictors of length of hospitalization in trauma patients. Can J Surg. 2005;48:361–6.PubMedGoogle Scholar
- Munoz E, Lory M, Josephson J, Goldstein J, Brewster J, Wise L. Pediatric patients, DRG hospital payment, and comorbidities. J pediatr. 1989;115:545–8.PubMedView ArticleGoogle Scholar
- Jencks SF, Williams DK, Kay TL. Assessing hospital-associated deaths from discharge data. JAMA. 1988;206:2240–6.View ArticleGoogle Scholar
- Munoz E, Rosner F, Friedman R, Sterman H, Goldstein J, Weis L. Financial risk, hospital cost, and complications and comorbidities in medical non-complications and comorbidity-stratified diagnosis-related groups. Am J Med. 1988;84:933–9.PubMedView ArticleGoogle Scholar
- Hsia DC, Krushat WM, Fagan AB, Tebbutt JA, Kusserow RP. Accuracy of diagnostic coding for Medicare patients under the prospective-payment system. N Engl J Med. 1988;318:352–5.PubMedView ArticleGoogle Scholar
- Green J, Wintfeld N. How accurate are hospital discharge data for evaluating effectiveness of care? Med Care. 1993;8:719–31.View ArticleGoogle Scholar
- Lorenzoni L, Cas RD, Aparo UL. Continuous training as a key to increase the accuracy of administrative data. J Eval Clin Pract. 2000;6:371–7.PubMedView ArticleGoogle Scholar
- Preyra C. Coding response to a case-mix measurement system based on multiple diagnosis. Health Serv Res. 2004;39:1027–45.PubMedView ArticleGoogle Scholar
- Matsuda S. Casemix as a tool of transparency of medical services. Jpn J Soc Sec Policy. 2007;6 Suppl 1:43–53.Google Scholar
- Iezzoni LI. Data sources, implications: information from medical records, patients. In: Iezzoni LI, editor. Risk adjustment for measuring healthcare outcomes. 2nd ed. Chicago: Health Administration Press; 1997. p. 243–78.Google Scholar
- Ishizaki T, Imanaka Y, Hirose M, Kuwabara K, Ogawa T, Harada Y. A first look at variations in use of breast conserving surgery at five teaching hospitals in Japan. Int J Qual Health Care. 2002;5:411–8.View ArticleGoogle Scholar
- OECD Health Data (2003) Show health expenditures at an all-time high. Available from URL: http://www.oecd.org/dataoecd/10/20/2789777.pdf/[Accessed 2007 Nov 14].
- Matsuda S, Imanaka Y, Kuwabara K, Fushimi K, Hashomoto H, Ishikawa BK. Japan case mix project-general perspective. In: Proceeding of the 18th International Case Mix Conference PCS/E 2002, Innsburuck, Austria, p. 722–6.Google Scholar
- Kuwabara K, Imanaka Y, Matsuda S, Fushimi K, Hashimoto H, Ishikawa KB. Profiling of resource use variation among six diseases treated at 82 Japanese special functioning hospitals, based on administrative database. Health Policy. 2006;78:306–18.PubMedView ArticleGoogle Scholar