Severe mental illness and health service utilisation for nonpsychiatric medical disorders: A systematic review and meta-analysis

Summary

Background

Psychiatric comorbidity is known to impact upon use of nonpsychiatric health services. The aim of this systematic review and meta-analysis was to assess the specific impact of severe mental illness (SMI) on the use of inpatient, emergency, and primary care services for nonpsychiatric medical disorders.

Methods and findings

PubMed, Web of Science, PsychINFO, EMBASE, and The Cochrane Library were searched for relevant studies up to October 2018. An updated search was carried out up to the end of February 2020. Studies were included if they assessed the impact of SMI on nonpsychiatric inpatient, emergency, and primary care service use in adults. Study designs eligible for review included observational cohort and case-control studies and randomised controlled trials. Random-effects meta-analyses of the effect of SMI on inpatient admissions, length of hospital stay, 30-day hospital readmission rates, and emergency department use were performed. This review protocol is registered in PROSPERO (CRD42019119516). Seventy-four studies were eligible for review. All were observational cohort or case-control studies carried out in high-income countries. Sample sizes ranged from 27 to 10,777,210. Study quality was assessed using the Newcastle-Ottawa Scale for observational studies. The majority of studies (n = 45) were deemed to be of good quality. Narrative analysis showed that SMI led to increases in use of inpatient, emergency, and primary care services. Meta-analyses revealed that patients with SMI were more likely to be admitted as nonpsychiatric inpatients (pooled odds ratio [OR] = 1.84, 95% confidence interval [CI] 1.21–2.80, p = 0.005, I² = 100%), had hospital stays that were increased by 0.59 days (pooled standardised mean difference = 0.59 days, 95% CI 0.36–0.83, p < 0.001, I² = 100%), were more likely to be readmitted to hospital within 30 days (pooled OR = 1.37, 95% CI 1.28–1.47, p < 0.001, I² = 83%), and were more likely to attend the emergency department (pooled OR = 1.97, 95% CI 1.41–2.76, p < 0.001, I² = 99%) compared to patients without SMI. Study limitations include considerable heterogeneity across studies, meaning that results of meta-analyses should be interpreted with caution, and the fact that it was not always possible to determine whether service use outcomes definitively excluded mental health treatment.

Conclusions

In this study, we found that SMI impacts significantly upon the use of nonpsychiatric health services. Illustrating and quantifying this helps to build a case for and guide the delivery of system-wide integration of mental and physical health services.

Keywords:

Critical care and emergency medicine – Hospitals – Diagnostic medicine – Psychoses – Schizophrenia – Inpatients – Bipolar disorder – diabetes mellitus

Introduction

Mental health conditions are associated with high disease burden, poor overall health outcomes, and high health service utilisation [1–4]. Arguably, increased health service utilisation could be attributed to appropriate use of psychological and psychiatric services. However, an early review (1994) of the literature found that psychiatric comorbidity (particularly cognitive and organic mental disorders) was associated with increased length of stay (LOS) in the general hospital [5]. Building on this, a later review (2005) assessed nonorganic common mental disorders and found that depression was associated with higher use of general medical services [6]. The most recent review (2018) found that medical inpatients with any psychiatric comorbidity had longer hospital stays, higher medical costs, and more readmissions than inpatients without [7].

Although severe mental illness (SMI) was not precluded from Jansen and colleagues’ review [7], the specific impact of SMI on nonpsychiatric health service use was not reviewed or quantified. People with SMI, such as schizophrenia or psychotic disorder, are more likely to develop chronic physical illness than the general population [8], and the impact of physical illness on people with SMI is significantly greater [9]. It is probable that this affects the use of nonpsychiatric general medical services in this patient group. Moreover, there are serious inequalities in the provision of physical healthcare for patients with SMI [8–10], which likely have repercussions for how they use general medical services.

Therefore, we sought to specifically review the literature surrounding the impact of SMI on the use of nonpsychiatric inpatient, emergency, and primary care services for patients with medical disorders. When possible, meta-analysis was used to determine the effect that SMI had on specific outcomes.

Methods

This review protocol is registered in the PROSPERO International Prospective Register of Systematic Reviews (https://www.crd.york.ac.uk/PROSPERO/) (CRD42019119516). The protocol conforms to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines; the relevant checklist is provided in S1 Appendix. This research is part of a larger systematic review assessing the impact of SMI and personality disorder on nonpsychiatric health service utilisation. The present study focused on the impact of SMI on inpatient, emergency, and primary care service use. Ethical approval was not required.

Search strategy and selection criteria

We searched PUBMED, Web of Science, PsycINFO, EMBASE, and The Cochrane Library for relevant studies with no publication date restrictions (see S2 Appendix for the full search strategy for each database searched). The search was supplemented with hand searches of journals related to the field and reference sections of relevant papers. Searches were carried out between 26 October 2018 and 2 November 2018. The search strategy was developed and conducted by AR. An updated search was carried out up to the end of February 2020.

Different definitions of SMI exist, but it generally refers to illnesses associated with psychosis as adopted by the Quality and Outcomes Framework of the United Kingdom (UK) National Health Service [11]. Therefore, in this review, SMI included bipolar disorder, psychosis, schizophrenia, and/or schizoaffective disorder. SMI did not include major depressive disorder (MDD). Studies that included MDD in their definition of SMI were excluded unless results were presented separately for each SMI subtype. Patients with physical health conditions and/or receiving medical treatment who did not have SMI served as controls.

Health service utilisation for medical nonpsychiatric disorders was the primary outcome. The current paper focused on inpatient (number of hospital admissions, likelihood of hospital admission, LOS in days, risk of longer LOS, number of readmissions, likelihood of readmissions), emergency (number of emergency department visits, likelihood of an emergency department visit), and primary care service use (number of primary care visits, likelihood of a primary care visit). We excluded studies for service use related to psychiatric, psychological, mental, or behavioural disorders. We included observational cohort and case-control studies and randomised controlled trials. We excluded reviews, case reports, and studies that used qualitative methods only.

For a study to be included, patients had to be 16 years or older. The majority of studies were explicit about excluding paediatric/adolescent patients. Where it was unstated whether all patients were over 16 years (four studies), the decision to include the study was based on the likelihood of the index medical condition occurring in a paediatric sample. For example, a sample of stroke patients will unlikely have a sizeable paediatric sample and would therefore be included, whereas a sample of patients undergoing trauma surgery could have a sizeable paediatric sample and would be excluded. All studies had to be published in peer-reviewed journals. Non-English-language articles were excluded. Conference proceedings were also excluded. When conference proceedings emerged in the search, authors were contacted to ascertain whether the data had been published in a peer-reviewed journal.

Articles were independently screened in two stages: a title and abstract screen (AR, AJ), followed by the retrieval and screening of potentially relevant full-text articles by two reviewers using the criteria listed above (AR, LE). Interrater reliability for the full-text screen was assessed using Cohen’s kappa, which indicated moderate and substantial levels of agreement between the reviewers for the original and updated search, respectively (original: κ 0.55, 77.6% agreement; updated: κ 0.64, 81.8% agreement). Conflicts were resolved through discussion.

Data extraction and quality assessment

Data were extracted by two reviewers (AR, SJ): AR extracted the data from the publications and SJ cross-checked 10% of the extracted studies for accuracy. There was acceptable agreement on extraction (intraclass correlation coefficient = 0.63). Sample characteristics, methodological characteristics, and main health service utilisation outcomes were extracted. The data extraction tables were piloted and refined before extraction began. As all studies included in the review were observational studies, the Newcastle-Ottawa Scale (NOS) was used to assess the quality of each study [12]. The NOS assesses the quality of each study using a system in which ‘stars’ are awarded on three broad categories: selection of groups, comparability of groups, and discernment of the outcome of interest for the case-control or cohort. Each article is rated on nine variables and can earn a maximum of nine stars. More stars indicate less risk of bias in a given study, and the number of stars awarded allows a study to be deemed of good, fair, or poor quality. Because of the nature of the current review, if a study did not adjust for severity of physical illness and/or the presence of physical comorbidities (e.g., Charlson Comorbidity Index [13], Elixhauser Comorbidity Index [14], a list of relevant physical comorbidities, a measure of physical illness severity), it was deemed to be of poor quality, regardless of the number of NOS ‘stars’ acquired. Quality assessment was carried out with the outcome of interest in mind; i.e., if a study had several clinical outcomes alongside a health service use outcome, the quality of the study would be assessed based on the health service use outcome.

Data analysis

For all outcomes (inpatient service use, emergency service use, and primary care use), a narrative synthesis was carried out.

Meta-analysis and subgroup analysis were performed using Review Manager 5.3 of the Cochrane Collaboration [15]. In all meta-analyses, we used a random-effects model, since this model estimates effects while considering the heterogeneity between studies. For studies that reported continuous data, only those that provided both means and standard deviations/standard errors were included in the meta-analysis. Where required, standard deviations were calculated from confidence intervals (CIs) using a verified formula [16]. For studies that reported the likelihood of an outcome occurring, only adjusted studies (physical comorbidities/illness severity + other relevant factors) that provided odds ratios (ORs) and 95% CIs were included in the meta-analysis. All ORs and CI limits were log-transformed (natural log). Standard error was then calculated using a verified formula [16].

Higgins’ I² was used to assess heterogeneity between studies. Where considerable heterogeneity is present (I² ≥ 90%) [16], statistical pooling is usually deemed inappropriate. However, we have reported pooled effect sizes for ease of interpretation, even in cases where there was considerable heterogeneity due to the clinical relevance of results. These should be interpreted with caution.

Sources of heterogeneity were investigated using subgroup analysis. Sources investigated included type of health service use outcome (all-cause [i.e., psychiatric treatment is unlikely but cannot be definitively ruled out] versus medical), type of SMI, sample size (cutoff determined by median split), country where the study took place, and study quality, when applicable. We assessed the degree of publication bias by visual examination of funnel plots. Publication bias was deemed to be absent if the plot showed an inverted symmetrical funnel. In all analyses, statistical significance was set at p ≤ 0.05.

Results

Study selection

The systematic literature search resulted in a total of 4,620 articles. After removing duplicates, 3,507 articles remained. Preliminary hand searches of journals and reference sections of relevant papers identified eight more articles. After reviewing the titles and abstracts of these 3,515 articles, a total of 196 articles were included for the full-text review. The updated search carried out in March 2020 resulted in a total of 347 articles. After reviewing the titles and abstracts of these 347 articles, 33 were included in the full-text review (see PRISMA diagram in Fig 1). A list of excluded articles is provided in S3 Appendix.

Of 87 eligible studies, 74 (initial search, 61 studies; updated search, 13 studies) entered the review, as these assessed the impact of SMI on inpatient, emergency, and/or primary care services. All were observational cohort or case-control studies, and most were retrospective cohorts in which health service utilisation was compared between patients with and without SMI over time (n = 63); some studies adopted matched case-control designs (n = 11). Forty-five studies were deemed to be of good quality, three were fair, and 26 were poor.

The number of participants in the reviewed studies ranged from 27 to 10,777,210. The majority of the studies were from the United States (US: n = 51), with the remainder in the UK (n = 7), Canada (n = 4), Denmark (n = 4), Taiwan (n = 3), Australia (n = 2), Israel (n = 1), Japan (n = 1), and Sweden (n = 1). The majority of studies were carried out in patients with specific medical index disorders (n = 54). The most common medical index disorders were diabetes (n = 10), heart failure (n = 7), and total joint (knee and/or hip) arthroplasty (n = 7). The remaining studies (n = 20) were carried out in the general medical population, e.g., patients admitted to general medicine departments [17,18], all residents of nursing homes in Florida [19], all people on the Taiwanese National Health Research Institute Database [20]. Study periods ranged from 3 months to 18 years.

The impact of SMI on the use of nonpsychiatric medical inpatient services

Inpatient hospital admissions

Table 1 describes the 27 studies (comprising 44 separate analyses) of the impact of SMI on nonpsychiatric medical inpatient admissions [19–45]. In 35 of these analyses, having SMI was associated with increased inpatient hospital admissions over the study period (12 months to 15.5 years). Seven analyses from five studies produced nonsignificant results [22,28,33,35,43] and two analyses revealed that major psychotic disorder and schizophrenia was associated with reduced inpatient admissions in residents of assisted living facilities under 65 years and veterans under 60 years of age, respectively [22,33].

Results from 17 analyses from nine studies were deemed potentially appropriate for inclusion in a meta-analysis [21,23,25,27,31,33,35,36,43]; i.e., they adjusted for physical comorbidities/illness severity and other relevant factors and presented ORs relating to the likelihood of inpatient hospitalisation over the study period. However, Higgins’ I² indicated that the set of studies were extremely heterogeneous (χ² [16] = 10,649.77, p < 0.001, I² = 100%); therefore, the estimation of the overall pooled effect should be interpreted with caution. The pooled OR indicated that patients with SMI were significantly more likely to be admitted as a nonpsychiatric inpatient than patients without SMI (pooled OR = 1.84, 95% CI 1.21–2.80, p = 0.005; Fig 2).

In order to determine the source of heterogeneity, we examined estimates between analyses that examined all-cause hospital admissions versus medical admissions, SMI subtype, sample size (determined by median split; n > 155,312 versus n ≤ 155,312), and analyses in the US versus Canada. Results showed that likelihood of inpatient admission did differ across SMI subtype (test for subgroup differences: χ²[2] = 11.35, p = 0.003) with patients with bipolar disorder having the lowest risk of admission (OR = 1.35, 95% CI 1.07–1.71, p = 0.01) and those with schizophrenia having the highest (OR = 2.37, 95% CI 1.09–5.15, p = 0.03). No other factors explained heterogeneity (see S4 Appendix for tables detailing subgroup analyses statistics). Study quality was not considered to be a potential source of heterogeneity, as all but one study included in the meta-analysis was of good quality. Visual examination of the funnel plot (see S5 Appendix) did not indicate significant publication bias.

Length of hospital stay

Table 2 describes the 30 studies (containing 38 separate analyses) that assessed the impact of SMI on nonpsychiatric LOS [17,18,20,28–30,32,40–42,46–65]. In 29 of the 38 analyses, SMI was associated with increased LOS. Eight analyses reported no significant associations [46,48,50,51,63,65], and one study found SMI associated with shorter hospital stays [29].

Fifteen studies (17 analyses) were potentially suitable for meta-analysis, as they reported the mean LOS (and standard deviation or CIs) for patients with and without SMI [20,28,30,41,42,46,48,51,54,56,58,59,61,62,65]. The Higgins’ I² indicated that the set of studies were extremely heterogeneous (χ²[16] = 15,432.12, p < 0.001, I² = 100%); therefore, the estimation of the overall pooled effect should be interpreted with caution. The pooled standardised mean difference (SMD) indicated that nonpsychiatric LOS was 0.59 days longer for patients with SMI compared to patients without (pooled SMD = 0.59 days, 95% CI 0.36–0.83, p < 0.001; Fig 3).

Sources of potential heterogeneity were investigated using subgroup analysis (see S4 Appendix). Results indicated that LOS differed across SMI subtypes (test for subgroup differences: χ²[3] = 15.73, p = 0.001), with patients with bipolar disorder having the shortest difference in LOS (SMD = 0.11 days, 95% CI −0.03 to 0.25) and those with schizophrenia having the highest (SMD = 0.86 days, 95% CI 0.50–1.21). There was also a difference in terms of study quality with significant differences between those with and without SMI emerging only in studies of poor quality. No other sources of heterogeneity emerged (see S4 Appendix). Visual examination of the funnel plot (see S5 Appendix) suggested publication bias.

Hospital readmission rates

Table 3 describes the 20 studies (36 separate analyses) included in the review that examined the impact of SMI on nonpsychiatric hospital readmission rates [26,64,66–83]. In 32 of these analyses, SMI was associated with increased hospital readmissions. Four studies showed no significant association [70,72,74,75].

Ten studies (16 analyses) were suitable for meta-analysis [67–69,73,75,77–80,83]. These studies examined the impact of SMI on 30-day readmission rates specifically, except for one study that looked at 28-day readmission rates [75]. Higgins’ I² (χ²[15] = 86.25, p < 0.001, I² = 83%) suggested that meta-analysis was appropriate [16]. The pooled OR indicated that patients with SMI were significantly more likely than patients without SMI to be readmitted to hospital within 30 days of the index medical hospitalisation (pooled OR = 1.37, 95% CI 1.28–1.47, p < 0.001; Fig 4).

Although meta-analysis was appropriate, there was still considerable heterogeneity between studies examining 30-day readmission rates. Geographical area (test for subgroup differences: χ²[1] = 13.89, p ≤ 0.001) was a significant sources of variance between studies (see S4 Appendix). Studies carried out in Europe (UK and Denmark) indicated that SMI patients were more likely to be readmitted to hospital within 30 days (OR = 1.65, 95% CI 1.48–1.85, p < 0.001) than their counterparts in the US (OR = 1.28, 95% CI 1.19–1.39, p < 0.001). No other sources of heterogeneity emerged. As all studies included in the meta-analysis were of good quality, study quality was not deemed to be a source of heterogeneity. Visual examination of the funnel plot (see S5 Appendix) indicated that publication bias was likely.

The impact of SMI on emergency department use for medical disorders

Table 4 describes the 15 studies (20 analyses) included in the review that assessed the impact of SMI on the use of emergency care [20,23,29,31,32,34,39,42,44,45,70,76,84–86]. In 18 out of 20 of the analyses, SMI was associated with increased use of emergency departments. This was true irrespective of adjustments for severity of medical disorders. Only two studies reported nonsignificant differences in the use of the emergency department between patients with and without SMI [34,70].

Meta-analysis was suitable for four studies (six analyses) [23,31,84,86]. The Higgins I² value indicated that there was extreme variation between studies (χ²[5] = 404.15, p < 0.001, I² = 99%); therefore, the estimation of the overall pooled effect should be interpreted with caution. The pooled OR indicated that patients with SMI were significantly more likely to attend the emergency department than patients without (pooled OR = 1.97, 95% CI 1.41–2.76, p < 0.001; Fig 5). Because of the small number of analyses included in the meta-analysis, subgroup analysis to determine sources of heterogeneity was not appropriate [16]. All studies were of good quality, meaning this did not contribute to the variation between them. Examination of the funnel plot indicated that publication bias was possible (see S5 Appendix).

The impact of comorbid SMI on use of primary care

Seven studies (10 separate analyses) looked at the impact of SMI on the use of primary care services (Table 5) [23,29,84,87–90]. Out of the 10 analyses, eight found that SMI was associated with increased primary care use. Five studies were of good quality and adjusted for physical illness severity, with the exception of that by Copeland and colleagues, who performed a cluster analysis [87], and Norgaard and colleagues, who provided unadjusted descriptive statistics [90]. One study found that there was no significant effect of SMI on primary care use in epilepsy patients [84]. Lichstein and colleagues reported that patients with schizophrenia were less likely to use medical homes for medical disorders (a model of primary care in the US) compared with patients without schizophrenia or depression [89]. Because of heterogeneity across outcomes, meta-analysis was not possible.

Discussion

This systematic review and meta-analysis aimed to understand the impact of SMI on use of general inpatient, emergency, and primary care services. The evidence showed that SMI leads to increased use of general medical services. More specifically, having an SMI is associated with increased inpatient admissions, increased length of hospital stay, higher 30-day readmission rates, more emergency room attendances, and increased use of primary care services, for nonpsychiatric reasons. The results of this review highlight the extent to which patients with SMI need targeted and effective interventions and system-wide integrated mental and physical healthcare.

The majority of studies indicated that nonpsychiatric inpatient admissions, LOS, hospital readmission rates, and emergency department use were increased in medical patients with SMI compared to patients without SMI. This was confirmed with meta-analyses that showed that patients with SMI were more likely to have an inpatient admission, had hospital stays that were increased by 0.59 days, and were more likely to be readmitted within 30 days compared to patients without SMI. Meta-analysis also showed that patients with SMI were more likely to attend the emergency department. Most studies included in this review also found that SMI was associated with increased use of primary care services. These findings are in line with previous reviews that have shown that health service utilisation is increased in patients with psychiatric comorbidity [5–7]. However, this is, to our knowledge, the first time that the impact of SMI on the use of general medical services has been systematically reviewed and meta-analysed. We believe these findings highlight the need for system-wide integration of mental and physical health services, particularly in secondary care. Although it is possible that increases in the use of primary care services associated with SMI are reflecting the provision of integrated care already in place, we do not believe this is the case for specialist secondary care services.

Subgroup analyses revealed some interesting findings surrounding factors that might impact upon how SMI affects health service utilisation. The likelihood of inpatient admission differed across SMI subtypes, with schizophrenia patients having the highest risk of admission and patients with bipolar disorder having the lowest. This mirrored results relating to LOS, which was longest in those with schizophrenia and shortest in those with bipolar disorder. Results relating to LOS should be interpreted with caution, however. Subgroup analysis revealed that only studies of poor quality showed that LOS differed significantly between those with and without SMI, which suggests that factors adjusted for in good-quality studies such as physical comorbidities and illness severity might better explain hospital LOS. Interestingly, the likelihood of 30-day readmission amongst patients with SMI was substantially reduced in studies carried out in the US. This perhaps reflects higher healthcare costs in the US [91], making patients less likely to seek readmission.

There are several factors that might explain why patients with SMI are using nonpsychiatric healthcare services more than those without. The primary reason for increased service use is likely the considerable rates of physical illness seen in patients with SMI. Significant increases in levels of obesity, metabolic syndrome, diabetes, cardiovascular disease, viral disease, respiratory tract disease, and musculoskeletal disease are seen alongside SMI, and illness severity is usually more pronounced in these patients [9]. Increased morbidity in SMI patients is largely down to a higher prevalence of modifiable risk factors [92], such as smoking [93], obesity [94], and alcohol and substance misuse [95]. Additionally, SMI is associated with physiological changes known to impact upon physical health. For example, changes in cortisol secretion associated with dysregulation of the hypothalamic pituitary adrenal (HPA) axis have been seen in patients with SMI [96,97]. Moreover, patients with SMI have increased levels of blood cytokines and circulating immune cell abnormalities [98,99], even at the early stages of mental illness [100]. Use of psychotropic medications has also been associated with increased obesity, dyslipidaemia, type 2 diabetes, and subsequent increased cardiovascular risk [101,102].

There is also evidence that disparities exist in the provision of healthcare for people with SMI, affecting the incidence and severity of physical disease. It is well documented that physical conditions in patients with SMI are underdiagnosed and suboptimally treated. SMI patients tend to have lower rates of medical and surgical intervention (e.g., cardiovascular stenting), and the quality of medical care, once received, can be substandard (e.g., levels of diabetes care) [10]. Moreover, the uptake of preventive strategies, such as cancer screening, is lower amongst those with SMI [103]. There are several reasons for these inequalities in health provision. Firstly, psychiatric symptoms may prevent the patient from seeking adequate physical healthcare. For example, cognitive impairment is often associated with SMI [104] and might impact upon a patients ability to access and understand health information and health services. Factors such as lack of motivation and self-neglect also often accompany SMI and will likely affect the extent to which a patient accesses medical services and adheres to medical advice [10]. Secondly, the stigma of SMI pervades all aspects of society, including healthcare. Stigmatisation of patients by physicians and other healthcare professionals can lead to diagnostic overshadowing and a lack of adequate care [105]. Diagnostic overshadowing might also be a result of a lack of training in physical or mental health or a lack of knowledge surrounding symptom recognition [106]. Moreover, it is possible that primary care physicians and psychiatrists feel unable to provide both physical and mental healthcare to SMI patients because of time constraints or lack of clinical training [10]. Thirdly, there is a known inverse association between socioeconomic status and mental illness [107]. Socioeconomic disadvantage is associated with poor access to healthcare. This association is likely to be stronger in some countries more than others, which is reflected in the results of the current study which showed that 30-day readmission is less likely for SMI patients in the US, where healthcare is costly.

All of these factors taken together increase the risk of treatment delay and the development of complications in patients with SMI. A lack of adequate physical healthcare for patients with SMI means that their physical symptoms are likely to be much worse when they finally present in general medical services, potentially leading to an increase in hospital admissions, rates of readmission, emergency department attendance, and use of primary care services, which we have described in this review.

Limitations

The majority of studies included in this review were of good quality with large sample sizes. In most cases, SMI was defined using established diagnostic codes, and health service utilisation outcomes were obtained using medical record linkage. However, some studies (35%) were rated as poor quality, largely owing to failure to control for physical illness severity or the presence of comorbidities, meaning the strength of evidence differed across studies. Nevertheless, most studies showed that SMI led to an increase in the use of general medical services, and study quality was not a significant source of heterogeneity in all but one meta-analysis. However, most studies included in the meta-analysis looking at LOS were of poor quality, which should be taken into account when interpreting the result. Only peer-reviewed publications were included in the review, meaning that publication bias was likely. Despite carrying out literature searches on several databases using comprehensive search strategies, it is possible that retrieval of all relevant research was not complete.

I² values indicated considerable heterogeneity across studies. This was to be expected in a review of this kind, in which there is large variation between studies in terms of patient population, hospital setting, and health system. Examining sources of variation revealed that certain outcomes differed according to SMI subtype and geographical location. Other patient and clinical characteristics likely explain the bulk of heterogeneity. This means that there is uncertainty around the magnitude of the impact of SMI on nonpsychiatric health service utilisation, and results of this review should be interpreted with this in mind.

In terms of the quality assessment of studies included in the review, the NOS has been criticised in terms of its subjectivity, and interrater reliability between authors and reviewers of studies has been found to differ significantly in that reviewers tend to view studies more favourably than the authors [108]. However, there is no gold standard when it comes to quality-assessment tools for observational studies [109]. Moreover, in the current review, we adopted quite stringent criteria for deciding when a study was poor quality (i.e., when a study did not adjust for severity of physical illness and/or the presence of physical comorbidities), meaning that the purported favourable view taken by reviewers might have been offset by that.

Although the focus of this review was on nonpsychiatric health service utilisation, it was not always possible to determine whether the health service outcome definitively excluded mental health treatment (e.g., use of the emergency department). Because of the nature of primary care, it was not possible to rule out the use of primary care services for psychiatric reasons in any of the studies included in the review. However, a recent study reported that nine out of 10 of the most common patient-reported reasons for primary care visits were nonpsychiatric [110], indicating that this outcome was likely to be mainly nonpsychiatric. It is very common for patients to have other psychiatric illnesses alongside SMI [111]. Unfortunately, because of the nature of the studies included, it was not within the scope of the current review to consider the impact of the overlap between SMI and other psychiatric conditions on nonpsychiatric health service utilisation. The studies included in this systematic review all described the impact of SMI on the use of general medical services in high-income countries, which affects the generalisability of the results. Future research should try to understand the impact of SMI, and more generally psychiatric comorbidity, on nonpsychiatric health service utilisation in low- and middle-income countries where health systems are not as well developed.

Conclusions

The results of this systematic review and meta-analysis highlight the extent to which SMI impacts upon general medical services. Patients with SMI were more likely to have an inpatient admission, had hospital stays that were increased by 0.59 days, were more likely to be readmitted within 30 days, and were more likely to attend the emergency department compared to patients without SMI. Most studies included in this review also showed that SMI was associated with increased use of primary care services. Illustrating and quantifying this helps to build a case for system-wide integration of mental and physical healthcare. Additionally, the results of the meta-analyses might be used to guide clinicians, policy makers, and commissioners in the improvement of the delivery of physical healthcare for SMI patients. The Five Year Forward View for Mental Health for National Health Service England aims to improve early detection of physical illness in SMI patients through the implementation of physical healthcare screening, assessment, and intervention [112]. This is to be delivered across both primary and secondary care. Prevention, early detection, and timely delivery of treatment for physical illness will hopefully improve the physical health of patients with SMI and reduce use of nonpsychiatric healthcare services.

Supporting information

S1 Appendix [doc]
PRISMA checklist for the systematic review and meta-analyses.

S2 Appendix [docx]
Literature search strategy.

S3 Appendix [docx]
List of excluded studies.

S4 Appendix [docx]
Subgroup analyses tables for meta-analyses.

S5 Appendix [docx]
Funnel plots for meta-analyses.