Health-economic outcomes in hospital patients with medical-psychiatric comorbidity: A systematic review and meta-analysis

Background Hospital inpatients often experience medical and psychiatric problems simultaneously. Although this implies a certain relationship between healthcare utilization and costs, this relationship has never been systematically reviewed. Objective The objective is to examine the extent to which medical-psychiatric comorbidities relate to health-economic outcomes in general and in different subgroups. If the relationship is significant, this would give additional reasons to facilitate the search for targeted and effective treatments for this complex population. Method A systematic review in Embase, Medline, Psycinfo, Cochrane, Web of Science and Google Scholar was performed up to August 2016 and included cross-references from included studies. Only peer-reviewed empirical studies examining the impact of inpatient medical-psychiatric comorbidities on three health-economic outcomes (length of stay (LOS), medical costs and rehospitalizations) were included. Study design was not an exclusion criterion, there were no restrictions on publication dates and patients included had to be over 18 years. The examined populations consisted of inpatients with medical-psychiatric comorbidities and controls. The controls were inpatients without a comorbid medical or psychiatric disorder. Non-English studies were excluded. Results From electronic literature databases, 3165 extracted articles were scrutinized on the basis of title and abstract. This resulted in a full-text review of 86 articles: 52 unique studies were included. The review showed that the presence of medical-psychiatric comorbidity was related to increased LOS, higher medical costs and more rehospitalizations. The meta-analysis revealed that patients with comorbid depression had an increased mean LOS of 4.38 days compared to patients without comorbidity (95% CI: 3.07 to 5.68, I2 = 31%). Conclusions Medical-psychiatric comorbidity is related to increased LOS, medical costs and rehospitalization; this is also shown for specific subgroups. This study had some limitations; namely, that the studies were very heterogenetic and, in some cases, of poor quality in terms of risk of bias. Nevertheless, the findings remain valid and justify the search for targeted and effective interventions for this complex population.


Introduction
Hospital inpatients often experience medical and psychiatric problems simultaneously. For patients who are admitted to hospitals for a general medical illness, the prevalence of a comorbid psychiatric disorder is estimated at 40% [1]. Conversely, psychiatric patients are at increased risk of developing comorbid general medical disorders [2]. The consequences of concurrent general medical and psychiatric illnesses include an increase in morbidity, mortality, healthcare utilization and costs [3]. All psychiatric disorders are associated with increased morbidity, and mortality and patients who suffer from eating disorders and substance dependence have the highest risk [4][5][6][7]. Numerous studies examined the relationship between medical-psychiatric comorbidities and health-economic outcomes. Most studies have evaluated the impact of comorbidities on three outcome measures: length of stay (LOS), medical costs and rehospitalization rates [8][9][10][11][12]. These parameters are considered major cost drivers in healthcare [13]. Although these studies suggest that medical-psychiatric comorbidities relate to health-economic outcomes, these outcomes have never been reviewed systematically. Furthermore, the nature of the relationship may be different for separate subgroups [8,11,14,15]. Meta-analyses on this subject have not been performed and, therefore, the overall pooled effect of the relationship between medical-psychiatric comorbidities and health-economic outcomes remains unknown.

Aims of the study
The aims of our study were as follows: 1) a systematic review of the literature to examine the relationship between medical-psychiatric comorbidity and health-economic outcomes; 2) an examination of possible differences in health-economic outcomes in patient subgroups; 3) an examination of the pooled-effect sizes for the effects of medical-psychiatric comorbidities on LOS, medical costs and rehospitalizations using meta-analyses. The importance of this review lies in its contribution to a cost-effective treatment of this complex and expensive population [16]. Policy makers might use the estimation of pooled-effect sizes for the effects of medicalpsychiatric comorbidity on the examined health-economic outcomes to improve the delivery of cost-effective care for patients with medical-psychiatric comorbidities. Furthermore, this review might stimulate future researchers to examine the impact of several subgroups with comorbidity on different health-economic outcomes more thoroughly. examined: Embase, Medline, Cochrane, Psycinfo, Web of Science and Google Scholar. The data extraction was performed on March 3, 2015 by a biomedical information specialist and updated on August 1, 2016: the search terms used are shown in Table 1 and an example of a full electronic search is provided in Table 2. The search terms were adjusted to comply with the specifications of the different databases. Duplicates and non-English articles were removed. A literature search based on cross-referencing was performed on November, 27, 2015. As no new articles were included after updating the search on August 1, 2016, no new cross-referencing was performed after this point.

Inclusion criteria
Empirical studies that examine the effect of having either medical or psychiatric comorbidity in hospital inpatients are included. These hospital inpatients have to be either primarily medically or psychiatrically ill and aged 18 years or older. Studies had to measure the effect on any of three major health economic outcomes (length of stay and/or medical costs and/or rehospitalization), and must be published in a peer-reviewed journal. Any measure of LOS (days, mean, median, et cetera), all types of medical costs (direct and indirect), and all reports of rehospitalization (all relevant time-frames) were included as the aim was to broadly review the literature on these outcomes. Inclusion criteria were limited to these outcomes since these are most reported in literature. It was hypothesized that medical-psychiatric comorbidity has a significant impact on these outcomes; it was further expected that effects were most prominent for inpatients, because these patients are presumed to be the most severely ill. By focusing on these inclusion criteria, we aimed to decrease heterogeneity between included studies. Consequently, study design was not an exclusion criterion, there were no restrictions on publication dates. Additionally, the examined populations had to consist of inpatients with medical-psychiatric comorbidities and controls. The controls were required to be patients without a comorbid medical or psychiatric disorder; however, the way the controls were sampled was not an exclusion criterion. Non-English articles were excluded.
Ideally, a comparison of comorbid patients (disease A and B) with patients that only have disease A or only disease B would be made. In this way, discerning whether the effect was additive or multiplicative could be estimated. However, only inpatients with disease A or B were included and compared with disease AB for reasons of feasibility.

Selection procedure
Two authors (LAWJ and MAvS) independently assessed 100 randomly selected titles and abstracts to validate the inclusion criteria. All obtained articles were then reviewed on the basis of title and abstract by the first author. Full texts of the included articles were then obtained. Subsequently, two authors (LAWJ and JAvW) independently assessed the selected articles in a standardized manner to further include or exclude articles for the review. Consensus was sought when disagreements between the authors on inclusion existed. When no consensus could be reached, the assessment of a third author (MAvS) was decisive. An intra-class correlation coefficient was calculated to examine the accuracy of match between the reviewers.
Thereafter, two authors (LAWJ and JAvW) read all included articles and extracted relevant data about nine predetermined characteristics: study design, patient characteristics, somatic diagnoses, psychiatric diagnoses, control group, moment of data collection (during or after treatment), LOS, medical costs, and rehospitalization rates. An electronic spreadsheet (Microsoft Excel) was used with these predetermined characteristics and available information for every included article was recorded. Again, disagreements on the collected data were resolved by discussion and, if no consensus was reached, the assessment of a third author (MAvS) was decisive.
The Newcastle-Ottawa Scale (NOS) for assessing the quality of non-randomized studies in meta-analyses was used to assess the risk of bias in each study [17]. The NOS scale has been developed to assess the quality of non-randomized studies with its design. It uses a "star system" that judges a study on three broad perspectives: selection of groups, comparability of groups, discernment of either the exposure or outcome of interest for case-control or cohort.
All included studies were rated by the first author (LAWJ) based on the NOS; using NOS, each article was rated on nine variables and could earn a maximum of "9 stars". More stars indicates less risk of bias in the assessed article. Finally, in reporting this review, the "Preferred reporting Items for Systematic reviews and Meta-Analyses" (PRISMA) checklist [18] was used. A review protocol was not used a priori to this systematic review.
Subgroups were selected based on the number of studies that researched the specific subgroup. Only the most extensive researched subgroups per health-economic outcome are presented in this study.

Meta-analyses
Meta-analyses were performed using Review Manager 5.3. In studies that reported continuous data, only those that stated means and standard deviations were included in the analyses. Furthermore, pooled-effect sizes were only reported if the (sub-)population was reasonable homogenous as this is not appropriate for heterogeneous studies [19]. The heterogeneity was based on statistics where a cut-off point of an I 2 of 50% was used. A random-effects model was applied to calculate treatment effects. In order to express treatment effects, the (standard) mean differences and 95% confidence intervals (CI) were calculated. The X 2 test was used to determine the heterogeneity between included studies where I 2 values of <25% represent low heterogeneity, between 25% and 50% represent moderate heterogeneity and values >50% suggest severe heterogeneity between the studies [20]. Statistical significance was assumed at P <0.05.

Study selection
The systematic literature search resulted in a total of 6163 articles. After removing duplicates and non-English articles, a total of 3165 studies remained. After reviewing all studies on title and abstract, a total of 86 articles were included for the full-text review. The reviewing authors agreed on the inclusion of 36 articles and the exclusion of 37 articles; an intra-correlation coefficient of 0.76 between the reviewers was estimated. Consensus was not reached for thirteen articles; after discussion, eight of these were included and five excluded. References in all 44 included articles were cross-checked. This resulted in ten extra articles to review in full text, eight of which were included. These studies did not appear in the search since the title did not refer to medical-psychiatric comorbidity but to a specific disorder. Finally, after the selection procedure, a total of 52 articles entered the definite literature review (" Fig 1").
A variety of study designs were included, mostly observational in nature. Longitudinal cohort study was the most common design. No randomized clinical trials were found.
All studies compared either a medical or psychiatric index disorder, 48 articles examined patients with a medical index disorder and a psychiatric comorbid disorder and, conversely, four articles examined atiens with a psychiatric index disorder and a medical comorbid disorder. The control groups consisted of inpatients without a medical or psychiatric comorbid disorder. The number of participants in the reviewed studies ranged from 63 to 1,617,710 patients. The impact of medical-psychiatric comorbidities on LOS was described in 42 (81%) articles, the impact on costs in 12 (23%) articles and on rehospitalization in ten (19%) articles. The risk of bias was assessed using the NOS and the scores ranged from the minimum (one star) to the maximum number (nine) of stars. Details of the individual studies are presented in Table 3, which is divided by outcome measure and sorted according to the NOS grading system. No additional risk of bias was performed: the risk of publication bias was low since both negative and positive findings were of interest for all measures (LOS, costs, rehospitalization).

Results of individual studies
Impact of medical-psychiatric comorbidities on LOS. Table 3 shows the results of the 42 included studies that examined the impact of medical-psychiatric comorbidities on LOS. Articles with the least risk of bias are shown at the top of the table and are arranged according to  Table 3. Impact of medical-psychiatric comorbidity on length of stay (LOS), medical cost and rehospitalization.
Impact of medical-psychiatric comorbidities on medical costs. Table 3 shows the articles that described the impact of medical-psychiatric comorbidities on medical costs. Out of 12 studies, nine (75%) showed a significant relationship [8,11,29,32,36,53,[58][59][60] and three (25%) showed a non-significant relationship [10,27,59].  The results of five (42%) studies were included in a meta-analysis " Fig 3". These studies were very heterogenous and, therefore, an estimate of an overall pooled effect was not applicable. However, "Fig 3" reveals that medical-psychiatric comorbidities are related to increased medical costs.
Impact of different subgroups on health-economic outcomes. In the reviewed studies, two subgroups-depression and delirium-appeared to be extensively studied and were suitable for further subgroup analysis.
Six studies provided suitable data for meta-analysis; the overall pooled mean difference in " Fig 4" showed that patients with comorbid depression had an LOS that was 4.38 days longer than patients without comorbid depression (mean LOS: 4.38 days; 95% CI: 3.07 to 5.68 days). The weighted average mean LOS of the depressed group was 13.8 days as opposed to 10.5 days for the non-depressed group.  Health-economic outcomes in patients with medical-psychiatric comorbidity Another extensively studied subgroup consisted of patients with comorbid delirium. Seven (16%) articles found a relationship between comorbid delirious patients and increased LOS compared to patients without comorbid delirium; five studies (71%) showed a significant relationship [32,37,42,45,52] and two studies (29%) a non-significant relationship [41,51]. The data in these articles were not sufficient to perform a meta-analysis.
Impact of comorbid depression on medical costs. The relationship between comorbid depression and medical costs was examined in six (46%) studies; of these, five (92%) showed significantly higher medical costs in comorbid depressed patients compared to patients without comorbid depression [29,36,53,58,60]. One (8%) study indicated a non-significant relationship [32]. A meta-analysis was not performed as only two articles provided limited data.
Impact of comorbid depression on rehospitalization. The relationship between medical-psychiatric comorbidities and rehospitalization was, again, extensively examined in the subgroup with comorbid depression compared to patients without comorbid depression. This was described explicitly in four (40%) reviewed articles. Two (50%) studies found that the odds ratio for rehospitalization was significantly higher in patients with comorbid depression [51,64]. Another study (25%) showed that patients with comorbid depression and a history of prior hospitalizations within six months were three times more likely to be rehospitalized within 90 days [61]. One (25%) study did not find a significant increase of the odds ratio for rehospitalization in patients with comorbid depression [63]. Meta-analysis for this subgroup was not possible because only one article provided useful data.

Discussion
To our knowledge this is the first review that examined the relationship between medical-psychiatric comorbidity and health-economic outcomes. Our analysis shows that hospital inpatients having medical-psychiatric comorbidities have a longer LOS, higher medical costs and more rehospitalizations. No randomized trials were found, however, and only one included study reached the highest standardized NOS quality score. The pooled-effect measures were often not appropriate to interpret since the available studies were very heterogeneous [19]. Nevertheless, the (standardized) mean differences of all meta-analyses indicate that medicalpsychiatric comorbidity is indeed related to increased health-economic outcomes. Moreover, the subgroup of depressed patients shows an increased mean LOS of 4.38 days compared to patients without depression: this outcome was moderately heterogeneous (I 2 = 31%) and was therefore considered appropriate.

Policy and clinical implications
Our systematic literature review elucidates the importance of medical-psychiatric comorbidities on health-economic outcomes. It is, consequently, disappointing to find that the quality of the included studies was mostly limited and that the heterogeneity of study samples was huge. Future studies on quality improvement strategies should therefore examine the impact of medical-psychiatric comorbidities on health-economic outcomes. This will help care providers and policy makers to organize care for patients with medical-psychiatric comorbidity in the most efficient way.
Our literature review suggests that the depressed subgroup of medical-psychiatric comorbid patients has stronger relationships with health-economics outcomes in comparison to nondepressed patients. For the first time in literature, we have established that, on average, hospital inpatients with comorbid depression stay in the hospital 4.38 days longer than non-depressed patients. In the Netherlands, the costs of a medical inpatient day range from €435 to €575 [66]. Based on the results of our review, the average medical costs in this country for patients with comorbid depression are increased to €1905-€2520 compared to patients without comorbidity. Researchers, care providers, policy makers and health insurers might use these outcomes for future research and healthcare policy making to improve cost-effective care for this subgroup. Furthermore, future prospective and randomized research should examine the impact of specific subgroups on health-economic outcomes more thoroughly to help hospitals improve cost-effective care for different subgroups with medical-psychiatric comorbidity.

Limitations
Although a thorough and extensive electronic search was performed on 6163 titles, this systematic literature review had several limitations.
First, the included studies were highly heterogenic regarding the patient population, type of hospital, country and year; thus, the results were difficult to compare, which reduced the possibility of investigating a pooled effect. Therefore, the magnitude of the impact on health-economic outcomes remains uncertain.
Ideally, patients with medical-psychiatric comorbidity were compared to studies that examined the medical and psychiatric illness separately. In this way, it might be possible to examine whether the effect of different combinations of medical and psychiatric illness was additive or multiplicative. For reasons of feasibility, our search strategy was narrowed to the impact of medical-psychiatric comorbidity on inpatients (as opposed to inpatients without psychiatric comorbidity) on health-economic outcomes. It was anticipated that including all possible combinations of general medical and psychiatric illnesses in the search would lead to an unmanageable number of included papers.
Second, in the past decades, the average LOS in hospitals has been largely reduced, which made it harder to compare studies over time. Publication date was not an exclusion criterion; hence, some studies were published more than two decades ago. As such, the results in these studies could reflect hospital-care patterns that have since changed.
Next, the search in the literature list led to the inclusion of eight extra titles. These titles were not found in the extensive search strategy since they described a specific disorder and the focus of the search was on medical-psychiatric comorbidity in general. While developing the electronic search strategy, it was explicitly decided to not include specific disorders since there was no clear cut-off point to determine when to stop including specific disorders in the search terms. Furthermore, if all disorders that exist in the literature were included in the search, the results of the literature search would be too broad.
Subsequently, the studies that examined medical costs had some methodological variation in terms of sampling and reporting those costs. Some studies examined the impact of total costs (inpatient, outpatient, pharmacy) and some only the in-hospital costs. Nevertheless, the evidence of almost all studies concerning the relationship between medical costs and medicalpsychiatric comorbidities pointed in the same direction. Consequently, it was suggested that comorbidity had an increasing effect on medical costs.
Moreover, the quality of the reviewed studies was variable and only one study received the full nine stars on the NOS. [17] Thus, the strength of the evidence differed among the three health-economic outcomes. Most studies examined the impact of medical-psychiatric comorbidities on LOS. As 95% of these studies showed an increased LOS, this relationship with medical-psychiatric comorbidities seemed the most reliable.
Another limitation of our study was that only one author rated the included articles. However, since almost all individual articles found results that pointed in the same direction, the risk of bias in the studies did not seem to impact the overall findings of this review. Additionally, there is a possible publication bias since only peer-reviewed articles were included. A study protocol was not developed in advance of this systematic review; it would therefore be challenging to note deviations and to assess if the outcomes of the review are reported according to the original study plan.
Finally, two subgroups of patients (comorbid depression and delirium) were comprehensively researched and therefore the results of studies that described these subgroups were used in the meta-analysis. This analysis was not predetermined but carried out post hoc since these groups appeared to be the most extensively studied in the included papers. Since the search was not focused on these subgroups, some literature that describes the impact of depression and a somatic comorbidity on health-economic outcomes may have been overlooked. Nevertheless, future researchers can use the results of this research to examine the impact of subgroups with a specific psychiatric disorder and a medical comorbidity.
Despite these limitations, the main conclusions of this review remain valid and clearly indicate, firstly, that general hospital inpatients with medical-psychiatric comorbidities showed longer LOS, higher medical costs and more rehospitalizations and, secondly, that the subgroup of depressed patients showed an increased mean LOS of 4.38 days compared to patients without depression.

Conclusion
This systematic literature review found a relationship between medical-psychiatric comorbidities and health-economic outcomes (LOS, medical costs and rehospitalizations). The metaanalysis for the subgroup depression showed an increased LOS (on average 4.38 days longer inpatient stay) compared to patients without depression. These results demonstrate that, on average, the medical costs for this subgroup are between €1905 and €2550 higher than patients without a comorbid depression. Policy makers might use these results to improve cost-effective care for this subgroup. Based on our results, we suggest that future research should examine the impact of several subgroups with medical-psychiatric comorbidity on health-economic outcomes more thoroughly.
Supporting information S1