Determinants of psychogeriatric inpatient length of stay and direct medical costs: A 6-year longitudinal study using a national database in Taiwan



Aims:  This research examined factors related to the average length of hospital stay (LOS) and average direct medical costs (DMC) for 2291 psychogeriatric inpatients (aged 65 and over) admitted for the first time to a psychiatric ward in 2002.

Methods:  Hospitalization claim data of these inpatients were traced for the subsequent 6 years (2002–2007) from the dataset of Taiwan's National Health Insurance program. Analysis was carried out using the t-test, χ2-test and zero truncated Tobit regression.

Results:  Mean LOS and mean DMC were significantly different according to sex, psychiatric diagnosis, institution type, ownership type, and number of hospitalizations, but age was the exception. Both LOS and DMC exhibited downward U-shape for the number of hospitalizations. Factors significantly associated with longer LOS and higher DMC were: male sex; schizophrenic and delusional disorders (compared with dementia); and public institution (compared with private hospital). Compared with dementia, organic mental and anxiety disorders had significantly shorter LOS, and affective disorders had shorter LOS but higher DMC. Community and psychiatric hospitals (compared with general hospital) significantly influenced LOS but not DMC.

Conclusion:  Our results can be used as a reference for providers and policymakers to improve psychiatric care efficiency and carry out National Health Insurance financial reform for psychogeriatric inpatients.

THE ISSUES OF determination of length of hospital stay (LOS) and direct medical costs (DMC) play an important role in evaluating the healthcare efficiency and quality, as well as hospital resource utilization, for psychiatric care.1,2 Compared with other diseases, the LOS and DMC of mental illness for the geriatric population are more important and complex because of their characteristics of chronicity, greater levels of functional disability, and cognitive and social functioning impairments.3 Due to the continuous and dramatic rise in the percentage of aged persons (≥65 years old) in the total population, from 4.1% in 1980 to 10.7% in 2010, in Taiwan, increasing numbers of psychogeriatric inpatients are in need of care.4,5 Therefore, exploring the factors that influence the LOS and DMC for psychogeriatric care is important for reducing the costs associated with care of geropsychiatric patients in the National Health Insurance (NHI) system.

In the previous reports, factors affecting psychiatric patient LOS among mixed age populations are associated with institutional characteristics,6–8 health-care system,9,10 demographic factors,11,12 and psychiatrists' caseload.13 However, very few studies focus on the factors for geriatric population LOS.3,8,14

In addition, prior studies discussing the factors related to psychogeriatric inpatient DMC are limited to one specific mental illness, such as Alzheimer's disease,1,2,15,16 affective disorder,17,18 schizophrenic disorder,19,20 anxiety disorder,21–23 or comparison of the costs among different mental disorders.24 Few studies analyze the factors influencing psychogeriatric inpatient DMC using broader variables which include comparisons among various mental disorders, demographic factors, and institutional characteristics. We thus attempt to fill this gap by capturing the complexities of factors affecting psychogeriatric inpatient DMC and LOS.

Using a dynamic national dataset of psychogeriatric inpatients in 2002 and the subsequent 6 years (2002–2007) from Taiwan's NHI program, this study had three aims: to analyze the relation of institution type and type of psychiatric diagnosis to LOS (or DMC); to examine the characteristics of psychogeriatric inpatients and types of health facility that affect the LOS (or DMC); and to explore the relation between LOS (or DMC) and the number of hospitalizations. The empirical findings obtained from this study can be used as a reference for providers and policymakers to improve psychiatric care efficiency and carry out NHI financial reform for psychogeriatric inpatients.


Data source

This study used a unique dynamic dataset of psychogeriatric inpatients (aged 65 and over) in 2002 and the subsequent 6 years (2002–2007) from Taiwan's NHI program, which covers 97% of the inhabitants in Taiwan.25 All information provided by hospitals, such as patient demographics, details of drug information and medical provider characteristics, is combined into the National Health Insurance Research Database (NHIRD). The Psychiatric Inpatient Medical Claim (PIMC) dataset was extracted from NHIRD inpatient claims data using a diagnosis of mental disorders covering the years 2002–2007. The ICD-9-CM codes 290–319 indicate mental disorders. Patients who were admitted to psychiatric wards for the first time from 1996 to 2001 were excluded from the PIMC database. From the PIMC database, we extracted information such as numbers of psychiatric inpatients, hospitalization frequencies, demographics, LOS, diagnoses, DMC, and ownership and institution types of health facilities.

To understand the common factors influencing psychogeriatric inpatients' LOS and DMC, we selected psychogeriatric inpatients admitted for the first time to psychiatric wards in 2002 (1 January 2002–31 December 2002). We eliminated patients who were foreign or lacked basic information, such as sex or age. Ultimately, a total of 2291 inpatients were included in our investigation. We then traced the hospitalization claims data of these psychiatric inpatients for the subsequent 6 years, during the period 2002–2007, to examine the number of hospitalizations, LOS for each hospitalization, and DMC for each hospitalization for all repeat hospitalizations. We excluded the data beyond 31 December 2007.

The dataset we used was obtained from NHIRD provided by the Bureau of National Health Insurance, Department of Health, and managed by National Health Research Institutes (NHRI). The data contained de-linked information and was released after patient identities and hospitals were scrambled cryptographically to protect privacy. This research project was approved by NHRI in terms of ethical concerns.

Analytic procedures

The LOS and DMC for a psychogeriatric inpatient in 2002 were outcome variables in our study. LOS (measured in days) indicates the average length of hospitalization stay which equals total length of stay over the total number of hospitalizations during the period 2002–2007. DMC (measured in $US) is the average direct medical costs, which is total hospitalization costs over the total number of hospitalizations during the same period. DMC includes drug expenses, service fee, acute bed fee and chronic bed fee, etc. The NHI paid all medical expenses during the time a patient stayed in a health facility.

There are two categories of explanatory variables in this study. The first is characteristics of psychogeriatric inpatients, such as sex, age, disease, and the number of hospitalizations (2002–2007). The second refers to type of health facility. Five-year intervals were used to divide psychogeriatric inpatients into six age groups: 65–69, 70–74, 75–79, 80–84, 85–89, and 90 years and older.7 As the top six diagnoses included more than 70% of psychogeriatric inpatients admitted for one major diagnosis (Table 1), our study included eight categories of diagnosis as follows: 290, dementia disorders; 294, organic mental disorders; 295, schizophrenic disorders; 296, affective disorders; 297, delusional disorders; 300, anxiety disorders; others (including ICD-9-CM codes 291–293, 298, 299, and 301–315); and more than one major diagnosis.

Table 1. Characteristics of psychogeriatric inpatients, types of health facilities, and univariate analysis of LOS and DMC in 2002 (n = 2291, 2002–2007)
Discrete characteristicsInpatientsLOSDMC
n %MeanSD P-valueMeanSD P-value
  1. P-values were obtained from t-tests or one-way anova as appropriate. Direct medical costs are in $US ($US1 = $NT31.42). Post-hoc (Bonferroni test) comparison with superscript.

  2. Dem < Sch, Dem > Anx, Dem > Oth, Org < Sch, Sch > Aff, Sch > Del, Sch > Anx, Sch > Oth, GH < CH, GH < PH; Dem < Sch, Dem > Anx, Dem > Oth, Org < Sch, Sch > Aff, Sch > Del, Sch > Anx, Sch > Oth, Org > Anx, Aff > Anx, Aff > Oth. (P < 0.05).

  3. DMC, direct medical costs; LOS, length of hospital stay.

All  55.77116.86 24393252 
Sex    0.0075  0.0038
 Female81935.7547.03109.89 21763018 
 Male147264.2560.64120.33 25863367 
Age (years)    0.8643  0.4881
 65–6948321.0857.69152.26 25824065 
 70–7459325.8850.2195.99 22412583 
 75–7960126.2357.16111.12 25233357 
 80–8439417.2058.01111.90 24133160 
 85–891777.7359.59105.88 25812791 
 ≥90431.8855.2886.59 20752148 
Psychiatric diagnosis        
 One major diagnosis186481.3649.77108.43<0.000122112987<0.0001
  Dementia (Dem)76933.5757.66114.94 22662999 
  Organic mental (Org)1365.9452.76117.29 21913235 
  Schizophrenic (Sch)1436.2496.91191.72 36995168 
  Affective (Aff)45519.8641.9387.53 22462645 
  Delusional (Del)642.7934.5828.08 21321577 
  Anxiety (Anx)823.5814.3927.38 8721029 
  Others (Oth)2159.3822.9144.62 15001546 
 More than one major diagnosis42718.6481.98145.41<0.000134364075<0.0001
Institution type        
 One institution type206990.3153.54118.28<0.0001237332800.088
  General Hospital (GH)145063.2946.32105.91 22712961 
  Community Hospital (CH)2018.7778.38139.97 27064409 
  Psychiatric Hospital (PH)41818.2566.64142.68 25633669 
 More than one institution type2229.6976.58100.580.0052306329050.0026
Ownership type        
 Only one ownership type217895.0755.54118.680.1336228732900.0002
  Public157468.7049.38134.27 20043398 
  Private60426.3657.90112.07 25873235 
 More than one ownership type1134.9360.2373.820.6777227223820.3533

In our regression analysis, the reference groups for patient characteristics were female sex, 65–69 years of age, and with diagnosis of dementia (ICD-9-CM code 290).

Hospitals were separated in accordance with institution and ownership types, respectively. Based on institution type, hospitals were divided into three categories: general hospital, community hospital and psychiatric hospital. Psychiatric hospitals were defined as those with beds exclusively for psychiatric patients. This condition did not exist in general hospitals or community hospitals. Hospitals were also classified by ownership type into private hospitals and public hospitals.

In order to reveal the effects of different hospital types, we extracted a group of inpatients admitted to two different institutional types or ownership types of hospitals in 2002.24 Since the numbers of psychogeriatric inpatients in general hospitals and public hospitals were larger than those in other institution types and private hospitals, respectively, we used the psychogeriatric inpatients that only stayed in general hospitals and public hospitals as a reference group for our research.

Statistical analysis

One-way anova and t-test analyses were used to examine the relation between LOS (or DMC) and the characteristics of psychogeriatric inpatients, types of health facilities, and the numbers of hospitalizations (2002–2007). To test the cross relations among LOS (or DMC), the type of health facility and diagnosis, we chose the top four diagnoses among all samples and used one-way anova.

During the data processing, LOS was positively skewed and DMC was so large that it biased our regression results. All outcome variables were given a natural logarithmic transformation that solved possible biases. Finally, zero truncated Tobit regression analysis was employed to assess the correlation between explanatory variables and outcome variables, since zero or negative LOS and DMC of patients were not included in our samples.


Descriptive analysis

In Table 1, the second column displays the characteristics of the psychogeriatric inpatients and types of health facilities, including sex, age, major psychiatric diagnosis, institution type of hospitalization, and hospital ownership. Of the 2291 patients admitted for the first time to a psychiatric ward in 2002, 819 (35.75%) were female and 1472 (64.25%) were male. Around 70% of patients were between 65 and 79 years old. Dementia disorder was the most prevalent category of psychiatric diagnosis at discharge (769, 33.57%). The second most prevalent category was affective disorders (455, 19.86%). Most patients were admitted to a medical center or the psychiatric division of a general hospital (1450, 63.29%), followed by psychiatric hospital (418, 18.25%). Patients had a relatively higher tendency to be cared for at a public hospital (1574, 68.70%).

Univariate analysis

In Table 1, the third and fourth columns present the t-test results of LOS and DMC for the characteristics of psychogeriatric inpatients and types of health facilities, respectively. For all samples, the mean LOS was 55.77 days (SD 116.86) with a range of 1–1953 days and the mean DMC was $US2439 (SD 3252) with a range of $US32–44 831. All of the variables were significant at P ≦ 0.05, except age (P = 0.8643, LOS; P = 0.4881, DMC) and hospital ownership type (P = 0.1336, LOS).

The relations of institution type and type of psychiatric diagnosis to LOS and DMC were analyzed and the results are displayed in panel A and panel B of Table 2, respectively. Four diagnoses (290, 294, 295, and 296) were analyzed. The mean LOS of schizophrenic disorders was the highest among these four categories regardless of the type of health facilities. These four categories of psychiatric illness showed significance at P ≦ 0.05 for general and psychiatric hospitals (Table 2, panel A). In addition, among institution types, the LOS were significantly different for both schizophrenic (P = 0.0354) and affective (P = 0.0863) disorders. For three institution types, there only existed significant differences in DMC among the four diagnoses for psychiatric hospitals (P = 0.0001).

Table 2. The relation among types of institution and psychiatric diagnoses in the LOS, DMC, and medical costs per day (n = 1397)
Diagnosis /Hospital typeGeneral hospitalCommunity hospitalPsychiatric hospitalMean P-value
(n = 952)(n = 131)(n = 314)
  1. DMC, direct medical costs; LOS, length of hospital stay; NP, number of patients.

Panel A: LOS
Dementia (NP = 703)53.97117.0882.17133.8659.22116.4758.30118.960.1462
Organic mental (NP = 131)42.84122.5344.7874.6071.13137.3751.66119.330.4782
Schizophrenic (NP = 127)57.97120.3087.5046.79152.39271.0899.05198.490.0354
Affective (NP = 436)34.9758.8031.1845.2656.30131.2238.2274.760.0863
P-value0.0440 0.2029 0.0019    
Panel B: DMC
Dementia (NP = 703)228331962675385820382248227531040.3313
Organic mental (NP = 131)224533841746227223083727216133010.7614
Schizophrenic (NP = 127)289941863045123648556732373553870.1354
Affective (NP = 436)212918341227136224354095213723310.1238
P-value0.2503 0.2457 0.0001    
Panel C: Medical costs per day
Dementia (NP = 703)96.6786.1147.1334.6271.4646.0085.6174.67<0.0001
Organic mental (NP = 131)83.5742.4562.7332.2564.8340.5973.6840.020.0216
Schizophrenic (NP = 127)68.7127.8438.4314.7557.3535.0962.9230.930.0387
Affective (NP = 436)84.3139.6183.6573.1967.1128.6481.5240.200.0050
P-value0.0021 0.0047 0.1694    

As DMC is related to the number of hospitalization days, which is also affected by the type of heath facility, we further measured the DMC per day (which is total hospitalization costs over the total number of days of hospitalizations). We examined the relations among types of institutions and psychiatric diagnoses in the DMC per day for an inpatient during the period 2002–2007, as shown in Table 2, panel C. From the comparisons of data in panel B and panel C, among four categories of psychiatric illness, DMC (measured per hospitalization) for schizophrenic disorders was the highest. In contrast, DMC was higher for dementia disorders when the average medical costs were measured per day, regardless of the types of institutions. Among the three types of hospitals, the mean DMC per day was the highest for general hospitals for the four categories of psychiatric illness, but DMC varied by diagnosis. Finally, the four categories of psychiatric illness showed significance at P ≦ 0.05 for all types of health facilities, however, there was no evidence of differences in DMC among institution types.

It is very common for psychiatric patients to have repeated hospitalizations. Thus, it is interesting to explore the relation between LOS (or DMC) and the number of hospitalizations. We first divided patients into 10 categories based on the number of hospitalizations as the top nine categories (from 1 to 9) included almost 99% of all samples (Table 3). Only one hospitalization is the most common among psychogeriatric inpatients (1485, 64.82%). Table 3 illustrates significant differences in LOS and DMC for all numbers of hospitalizations (P < 0.05). LOS was the longest for patients with five hospitalizations, while DMC was the highest for patients with eight hospitalizations. We show the results of Table 3 in Figure 1. Figure 1 exhibited a downward U-shape, suggesting that LOS and DMC increase at a low level of hospitalizations, but once a high enough level of hospitalizations is achieved, both values decrease with hospitalizations.

Table 3. Linear regression results for log(LOS) and log(DMC) (n = 2291)
EstimateSE P-valueEstimateSE P-value
  1. log(LOS), log-transformed mean of length of stay; log(DMC), log-transformed mean of direct medical costs.

Sex: Male (reference)      
Age: 65–69 years (reference)      
Psychiatric diagnosis: Dementia (reference)      
 Organic mental−0.08090.11190.47010.07610.09160.4059
 More than one major diagnosis0.26180.08030.00110.41440.0668<0.0001
Institution type: General hospital (reference)      
 Community hospital0.29100.09140.0014−0.09120.07470.2224
 Psychiatric hospital0.18090.06730.00720.02570.05500.6404
 More than one institution type0.16060.09890.10440.04220.08280.6106
Ownership type: Public (reference)      
 More than one ownership type−0.36260.12900.0049−0.30670.10790.0045
Number of hospitalizations0.28120.0329<0.00010.19290.0271<0.0001
(Number of hospitalizations)2−0.01220.0021<0.0001−0.00830.0016<0.0001
Figure 1.

The patterns of distributions of length of stay (LOS, in days) and direct medical cost (DMC, $US) along with the number of hospitalizations are downward-U shaped. inline image, LOS; inline image, DMC.

Linear regression results

We used linear regression to measure the contributions of each independent variable to explain variance in LOS or DMC, the dependent variable, submitted to a logarithmic transformation to correct the non-normal, significantly positive skew. Explanatory variables include demographic characteristics, diagnostic characteristics, hospital characteristics, and number of hospitalizations. Table 3 presents the linear regression results of LOS and DMC on these explanatory variables. As the dependent variable is treated as a log transformation to correct for the skew of the data, the coefficients themselves are not directly interpretable, though the signs on the coefficients are. The LOS was significantly higher for men than for women (P = 0.0008). The same was true for DMC (P = 0.0038). However, age factor appeared to be insignificant in explaining LOS or DMC differences.

When compared with dementia disorders, the LOS was significantly higher for diagnoses of schizophrenic disorders (P = 0.0002) and the DMC was significantly higher for diagnoses of schizophrenic disorders (P < 0.0001), affective disorders (P = 0.0075), and delusional disorders (P = 0.0828). However, both LOS and DMC were significantly lower for diagnoses of anxiety disorders (P < 0.0001).

Regarding the type of health facility, both community and psychiatric institutions carry significantly positive signs on their coefficients of LOS meaning an increase in a patient's LOS for these two types of institutions when compared with general hospitals (community, P = 0.0013; psychiatric, P = 0.0065). In addition, there were significant differences in DMC between private hospitals and public hospitals (P < 0.0001). The same did not hold true for LOS (P = 0.0182).

Finally, the downward U-shape seen in Figure 1 is supported by the linear regression results. A significantly positive sign on the coefficient of number of hospitalizations at P < 0.0001 level indicates that the larger the number, the higher the LOS or DMC. However, a negative sign of the coefficient of the square of the number of hospitalizations implies that the LOS and DMC decrease after reaching a critical number of hospitalizations.


Demographic factors

Our t-test results showed that the mean LOS for men (60.64 days) is significantly longer than that for women (47.03 days) at P = 0.0075, and the mean DMC ($2586) for men is also higher than that for women ($2176) at P = 0.0038. However, age had no significant effect on LOS or DMC. We obtained consistent results with a linear regression model. These findings are also consistent with those from most Asian countries, such as China,2 Japan,6 and Korea,7,8 but inconsistent with those from the USA.3

The higher LOS and DMC for male inpatients than for female inpatients might be caused by more aggressive behavioral symptoms26 and less adherence to treatment among men.27,28 Furthermore, elderly patients' condition might be influenced by filial piety, an important Confucian cultural element, and the shame and negative attitude toward mental illness.29 Chinese people are tolerant of their elders' psychiatric symptoms because of traditional beliefs and norms that elders must be treated with respect, protected from bad diagnostic news, and that harmony should be preserved within families.30 During the onset of the disease, families are more likely to keep the diagnosis a family secret31 and take care of the elder suffering from mental illness at home as part of their family obligation.32 As a result, treatment is typically only sought after the disease has become severe and requires hospitalization.

Types of mental illness

In general, elderly patients with psychiatric diagnoses can be assigned to two groups: minor psychiatric disorders (ICD-9-CM Codes 300–302, 306–311, and 316) and major psychiatric disorders (ICD-9-CM Codes 290–298).33 The LOS and DMC are significantly different between minor and major psychiatric diagnoses.8,33,34 Patients with minor psychiatric disorders (such as anxiety disorders) had lower inpatient health-care utilization because of their higher level of functioning and lesser reality distortion than those with major psychiatric disorders (such as schizophrenia). In other words, LOS and DMC could be predicted by mental illness diagnosis and level of functioning.

Our results imply that the types of mental disorders are important determinants of LOS and DMC. The heterogeneities and complexities among geriatric mental disorders lead to different levels of health-care utilization. LOS of dementia, schizophrenic, and organic mental disorders were longer than for other mental disorders (Table 1). Compared with dementia, schizophrenic disorders had significantly higher LOS and DMC but anxiety disorders had significantly lower LOS and DMC. In contrast, affective and delusional disorders had significantly higher DMC than dementia disorders.

Institutional factors

The type of institution also plays an important role in determination of inpatient LOS and DMC.7 We found that the LOS in general hospitals was significantly shorter than in community hospitals and psychiatric institutions. The differences in LOS between general hospitals and community hospitals were partly due to LOS control or different levels of medical and paramedical professional staff. In our dataset, psychiatrists' workload (bed/doctor ratio) was 9.27, 23.81, and 36.90 in general, community, and psychiatric hospitals, respectively. Thus, the various degrees of physician staffing and paramedical professionals in different types of institutions resulted in differences in inpatient LOS among hospital types.

In addition, LOS and DMC for private hospitals were significantly lower than for public hospitals. The non-price competition of hospitals under the NHI payment system forces hospital managers to improve the efficiency of medical care and to reduce excess services and operating costs. Institutions of different ownership types use different strategies for economic pursuit or to reflect shortages in health-care human resources in private hospitals.

Finally, legal actions that are taken by patients and their families for unsatisfying outcomes, malpractice or medical errors against physicians have increased rapidly in recent decades.35 It is reported that malpractice events in 2005 had more than triple the risk of those in 1991 (12.4% in 2005 vs 4.1% in 1991) to be sued in both civil and criminal courts in Taiwan.36 Under these circumstances, physicians in both private and public hospitals tend to defend themselves by applying more medical procedures or treatments to prevent lawsuits. Those actions indirectly cause unnecessary medical waste and increased LOS and DMC. The above different forces press health-care management executives and physicians to find the optimal care; that is, to maintain good medical quality and low cost simultaneously.


In this study, the data were obtained from the NHIRD of Taiwan's NHI program. There is no detailed individual private information, such as personal income, occupation, education level, race, marital status, lifestyles, preferences, symptoms and signs, and severity of illness, etc. Thus, we were unable to analyze the effects of these factors on LOS and DMC. Psychogeriatric inpatients' physical comorbidities (hypertension, diabetes mellitus, or chronic heart diseases, etc.) might have effects on the results of LOS and DMC. The investigation of those factors can be the future direction for research on factors related to LOS and DMC in psychogeriatric inpatients.


This study is based in part on data from the National Health Insurance Research Database provided by the Bureau of National Health Insurance, Department of Health and managed by the National Health Research Institutes. The interpretation and conclusions contained herein do not represent those of the Bureau of National Health Insurance, Department of Health or the National Health Research Institutes. All authors declare that they have no conflicts of interest.