Schizophrenia spectrum disorders (SSDs) are among the top 10 causes of disability worldwide [Reference WHO1]. In addition to causing disabling psychiatric symptoms, schizophrenia is associated with frequent physical illness [Reference Nasrallah, Harvey, Casey, Csoboth, Hudson and Julian2,Reference Saha, Chant and McGrath3]. People diagnosed with psychosis have a 2.5 times higher risk of dying than the general population [Reference Saha, Chant and McGrath3,Reference Walker, McGee and Druss4]. Mortality rates in the population with psychiatric disorders and the difference in mortality between patients diagnosed with psychosis and the general population has been increasing [3–Reference Björkenstam, Ljung, Burström, Mittendorfer-Rutz, Hallqvist and Weitoft5]. Psychiatric patients die earlier of similar causes than do the general population (e.g., heart disease, cancer, and cerebrovascular and respiratory diseases), and these patients account for more than two thirds of this excess mortality [3–Reference Kim, Chang, Bae, Cho, Lee and Kim9].
The adverse effects of psychiatric pharmacotherapy, high prevalence of unhealthy lifestyle and modifiable risk factors further increase the risks. Schizophrenia is, however, increasingly recognized as a systemic disorder and these patients face an additional burden in terms of somatic comorbidity implying overlapping and interacting disease mechanisms that involve neurotransmitter, inflammatory, endothelial and hormonal pathways among others [10–Reference Abraham, Miller, Birgenheir, Lai and Kilbourne13]. Growing body of evidence has demonstrated that people with psychosis are at greatly increased risk of chronic physical comorbidities (cardiovascular disease, metabolic syndrome, diabetes, and respiratory disease) [Reference De Hert, Correll, Bobes, Cetkovich-Bakmas, Cohen and Asai10,Reference Gardner-Sood, Lally, Smith, Atakan, Ismail and Greenwood11]. Furthermore, somatic comorbidities in psychiatric patients may be associated with the poor health related quality of life, independently of different sociodemographic, vital and clinical factors [Reference Filipčić, Filipčić IŠ, Matić, Lovretić, Ivezić and Bajić Ž12]. Also, it has been established that somatic comorbidities, as well as the negative association of somatic comorbidities with patients’ quality of life, are related to pharmacological treatment of psychosis [12–Reference Miller, Abraham, Bajor, Lai, Kim and Nord15]. However, access to preventive interventions, the quality of detection, diagnosis, and adequate treatment are still lower in psychotic than in non-psychiatric patients [Reference Björkenstam, Ljung, Burström, Mittendorfer-Rutz, Hallqvist and Weitoft5,Reference Laursen, Nordentoft and Mortensen6,Reference De Hert, Correll, Bobes, Cetkovich-Bakmas, Cohen and Asai10,16–Reference Correll, Detraux, De Lepeleire and De Hert19]. The prevalent current approach still separates physical and mental health care [Reference Fleischhacker, Cetkovich-Bakmas, De Hert, Hennekens, Lambert and Leucht20,Reference Smith, Langan, McLean, Guthrie and Mercer21].
Despite these facts, less is known about the association of somatic comorbidities with the outcome of treatment of psychosis. Several studies addressed the problem from this “reverse perspective” [22–Reference Dixon, Postrado, Delahanty, Fischer and Lehman26], but we have not found conclusive evidence that somatic comorbidities affect symptom exacerbation of schizophrenia spectrum disorders nor the efficacy of their treatment. The objective of this study was to explore whether the number of chronic physical illnesses is associated with a poorer SSD treatment outcome indicated by higher rate of psychiatric rehospitalizations independently of psychiatric comorbidities, and other clinical and sociodemographic parameters.
2.1. Study design
This cross-sectional study enrolled patients during 2016 at Psychiatric Hospital Sveti Ivan, Zagreb, Croatia. The study was nested within the larger frame of a prospective cohort study named, “Somatic comorbidities in psychiatric patients (SCPP)”, which has an expected end date of June 2017. The main study protocol was registered at ClinicalTrials.gov (NCT02773108), and it was approved by the Ethics Committee of Psychiatric Hospital Sveti Ivan. Informed consent was obtained from all patients. The study complied with World Medical Association Declaration of Helsinki 2013 [Reference World Medical Association27].
2.2. Study population
The targeted population was patients diagnosed with SSDs (ICD-10; DSM-V) who were treated in a psychiatric hospital and achieved a stable therapeutic dosage. Inclusion criteria were ICD-10 (F20-F29); DSM-V: schizophrenia spectrum disorders, both genders, age ≥ 18 years, treated in a psychiatric hospital as inpatients or outpatients, and ability to answer the questionnaire. Exclusion criteria were acute suicidality, dementia, mental retardation, acute psychosis, and intoxication. We chose a consecutive sample of outpatients by the order of their arrival at the exam, and all patients who were hospitalized during the enrollment period.
2.3. Needed sample size
Power analysis was performed before the start of the enrolment as the component of power analysis performed for the main prospective cohort study. A sample size of 231 achieves 90% power to detect an R 2 ≥ 0.05 attributed to two independent variables: number of physical illnesses and number of psychiatric comorbidities, using a F-test with a significance level (α) of 0.05. The independent variables tested will be adjusted for an additional 10 possible confounding variables with an R 2 ≥ 0.05. Expecting up to 15% of respondents would have missing data on dependent (number of hospitalizations) and independent variables, the initially needed sample size was determined to be n = 272. Power analysis was done in PASS 14 Power Analysis and Sample Size Software (2015) (NCSS, LLC, Kaysville, Utah, USA).
Our outcome was the number of psychiatric hospitalizations since diagnosis of the primary psychiatric illness. This is not direct measure of clinical success, however, it has been used in many observational studies as an outcome measure to evaluate antipsychotic effectiveness (e.g. [28–Reference Herceg, Jukić, Vidović, Erdeljić, Celić and Kozumplik31]) and, as Burns concluded, it showed as a good proxy outcome measure in schizophrenia [Reference Burns32]. The number of rehospitalizations was assessed objectively from hospital records archived in Psychiatric Hospital Sveti Ivan. We used the row number and not the standardized one (e.g., average number of hospitalizations annually) because we planned to control duration of illness since diagnosis and patients’ age by multivariate analysis.
2.5. Independent variables (predictors)
Our independent variables were number of chronic somatic comorbidities and the number of psychiatric comorbidities. Chronic somatic illness was defined as the non-mental illness that requires medical treatment and lasted for at least six months. Trained psychiatrists recorded all comorbidities after consultation of medical records and clinical interview.
2.6. Possible confounders
Possible confounders whose effect we tried to control by multivariate analysis were sex, age, education, marital status, number of household members, work status, diet, smoking, excessive alcohol consumption, physical activity, duration of primary psychiatric illness, severity at diagnosis measured by the Clinical Global Impression-Severity (CGI-S) scale, antipsychotic drugs used, and treatment with antidepressants and benzodiazepines. Alcohol consumption was measured by self-completion of 2nd wave European Health Interview Survey (EHIS) questions on frequency and number of standard alcoholic drinks . We calculated the average number of standard alcoholic drinks daily based on questions AL2. “Thinking of Monday to Thursday, on how many of these 4 days do you usually drink alcohol?” and AL3. “Number of alcoholic (standard) drinks on average on one of the days (Monday to Thursday)”, and the same two questions for the period Friday to Sunday (AL4 and AL5). Excessive alcohol consumption was defined as more than 20 g/day (2 standard units) for men and 10 g/day (1 standard unit) for women [Reference Catapano, Graham, De Backer, Wiklund, Chapman and Drexel34]. Duration and intensity of average PA was measured by EHIS-PAQ [Reference Finger, Tafforeau, Gisle, Oja, Ziese and Thelen35]. The instrument was developed for the second wave of European Health Interview Survey (EHIS). According to recent validation, it has good validity and reliability [Reference Baumeister, Ricci, Kohler, Fischer, Töpfer and Finger36]. EHIS-PAQ referent time is “average week”. Total PA defined as proportion of individuals being sufficiently physically active in total, that is who are performing ≥ 150 minutes of aerobic PA or ≥ 2 muscle-strengthening PA weekly [Reference Finger, Tafforeau, Gisle, Oja, Ziese and Thelen35].
2.7. Statistical analysis
The level of statistical significance was set at P < 0.05, and we gave all confidence intervals at 95% level. In all instances, we used two-tailed tests. According to the protocol, it was planned that in the case of > 5% of participants with missing data, multiple imputation would be done by fully conditional specification of the iterative Markov chain Monte Carlo method, and that we would do a sensitivity analysis. The main analysis was done by robust regression and iteratively reweighted least squares on all variables used simultaneously. We used Huber's method for robust influence function with Huber's default tuning constant of 1.345, and a default median absolute deviation of residuals’ scale factor of 0.6745. The criterion for stopping the iteration procedure was set at percent change of 0.001. Regression coefficients and tests of statistical significance were calculated assuming that the robust weights were random, calculated from the sample residuals, and not fixed. Multi-colinearity of independent and confounding variables was tested by tolerance, variance inflation factor (VIF), and eigenvalues/condition numbers. Independence of residuals was tested by the Durbin–Watson test. Normality of distributions was analyzed by Shapiro–Wilk and D’Agostino's omnibus K 2 tests. The model fit to the data was expressed by coefficient of determination (R 2) after robust weighting and by prediction error sum of squares (PRESS). Cases with missing data were omitted from multivariate analysis. No correction for multiple testing was needed, as all analyses were pre-planned, and only one multivariate analysis was interpreted. Sensitivity analysis was done in two ways. First, by repeating the analysis after multiple imputation of confounders’ missing data. Second, by repeating the analysis with particular antipsychotics replaced with a binary variable representing long-acting injections vs. oral therapy, and with particular antipsychotics replaced with a number of antipsychotics used. Statistical data analysis was done by NCSS 10 Statistical Software (2015) (NCSS, LLC. Kaysville, Utah, USA).
In the main, prospective cohort study, we included 1060 patients treated for any psychiatric condition (Fig. 1). The final sample of 301 patients diagnosed with SSDs was enrolled at the first measurement for the cohort study. Diagnoses of patients were determined by their psychiatrists, according to the ICD-10 diagnostic criteria (WHO, 1992). The percentages of their primary psychiatric diagnoses are showed in Table 1. Primary psychiatric illness duration ranged from < 1 year up to 49 years. Number of hospitalizations was mean (SD) 6.0 (7.61); median (interquartile range) 3.0 (1.0–8.0). Patients’ age ranged from 19 to 75 years (Table 2). The most prevalent chronic physical illnesses were endocrine, nutritional and metabolic diseases, and diseases of the circulatory and digestive system (Table 3). The most prevalent psychiatric comorbidities were disorders of adult personality and behavior, mental and behavioral disorders due to psychoactive substance use, mood (affective) disorders, and neurotic, stress-related, and somatoform disorders (Table 3).
Data are presented as number (percentage) of participants if not stated otherwise. : arithmetic mean; SD: standard deviation. Data were missing for duration of primary psychiatric illness 2 (0.7%), CGI-S 4 (1.3%) of participants.
a Sum exceeds 100% because of combination therapies.
Data are presented as number (percentage) of participants if not stated otherwise. : arithmetic mean; SD: standard deviation. Data were missing for marital status 1 (0.3%), number of household members 6 (2.0%), work status 3 (1.0), having a breakfast 7 (2.3), eating fruits 8 (2.7), eating vegetables 9 (3.0) of participants.
a Sufficient physical activity was defined as aerobic physical activity ≥ 150 min/week or ≥ 2 muscle-strengthening physical activity weekly.
b Excessive alcohol consumption was defined as > 20 g/day men; > 10 g/day women.
a Other diseases: I Certain infectious and parasitic diseases (A00-B99), VII Diseases of the eye and adnexa (H00-H59), VIII Diseases of the ear and mastoid process (H60-H95), XIX Injury, poisoning, and certain other consequences of external causes (S00-T98).
Final multivariate analysis was done per protocol with the list-wise deletion of cases with missing data on the sample of 267 (89%) patients with complete data. We had no missing data on the dependent nor on the independent variables. According to Little's MCAR test, our data were not missing completely at random (P = 0.001). Robust regression criterion for stopping the iteration procedure was achieved in 14 iterations. Multi-colinearity was not indicated, as tolerances of all variables were > 0.84, and variance inflation factors for all variables were < 3.1. The smallest eigenvalue was 0.15 with condition number of 22.4. The largest correlation was found between age and duration of illness (r = 0.48; r 2 = 0.23; P < 0.001). Distribution of residuals was symmetric and was not significantly different from the normal one: Shapiro–Wilk test, P = 0.092, D’Agostino Omnibus K 2 test, P = 0.085. The Durbin–Watson test statistic was 2.0, indicating no serial correlation of residuals. Several outliers were detected. Association of number of rehospitalizations with the number of physical illnesses and number of psychiatric diagnoses was not significantly deviated from linear (P = 0.260 and P = 0.714, respectively).
Univariate analysis showed significant positive correlation between the number of rehospitalizations and number of chronic physical illnesses (β = 0.20; P < 0.001). (Table 4, Fig.2). After adjustment for all clinical, sociodemographic and lifestyle characteristics by multivariate robust regression, the number of chronic physical illnesses was independently, significantly correlated with a larger number of psychiatric rehospitalizations (β = 0.16; P = 0.009) (Table 4). Having two or more chronic physical illnesses was associated with an increase of one psychiatric rehospitalization. The overall model was significant (P < 0.001) with coefficient of determination after robust weighting, R 2 = 0.54. The prediction error sum of squares (PRESS) coefficient of determination indicated low-to-moderate predictive validity (prediction R 2 = 0.15).
a Referent value was: up to once a week.
b Sufficient physical activity was defined as aerobic physical activity ≥ 150 min/week or ≥ 2 muscle-strengthening physical activity weekly.
c Excessive alcohol consumption was defined as > 20 g/day men; > 10 g/day women.
Sensitivity analysis after the multiple imputation of confounders’ missing data resulted in the comparable pooled independent association of chronic somatic illness with the number of psychiatric rehospitalizations (β = 0.15). Repeating the regression analysis with specific antipsychotics replaced with a binary variable representing long-acting injections vs. oral therapy, the relative importance of number of somatic illnesses further increased (β = 0.18). The same happened when we replaced specific drugs with the total number of antipsychotics used (β = 0.21).
Our study has shown significant and clinically relevant correlation of number of psychiatric rehospitalizations with the number of chronic physical illnesses in patients diagnosed with SSDs independently of psychiatric comorbidities, and other relevant clinical, sociodemographic, and lifestyle parameters.
Number of chronic physical illnesses may affect treatment outcome in different ways. First, more physically ill patients may have higher incidences of therapy discontinuation due to polypharmacy, increased risk of drugs interactions, and lower tolerability of antipsychotic medication [37–Reference Viktil, Blix, Moger and Reikvam39]. In addition, it is possible that patients sometimes erroneously recognize physical illness symptoms for the side effects of psychiatric therapy, which may jeopardize their adherence and lower efficacy, or somatic therapies’ adverse events may interfere with psychiatric treatment. Consequently, serious, chronic somatic illnesses may cause lower adherence and more frequent absence from psychotherapy sessions. Second, the activation of immune responses caused by physical illness may cause the worsening of the clinical picture [Reference Jakovljević and Ostojić40,Reference Kalakonda, Koppolu, Baroudi and Mishra41]. Third, capacity for productive participation in psychotherapy may be weaker because of somatic illnesses and subjective focus on their symptoms instead of on psychotherapy and treatment of the SSD. Fourth, chronic physical illnesses are associated with a higher probability of rehospitalization, what indicates higher relapse rates [Reference Boyer, Millier, Perthame, Aballea, Auquier and Toumi42,Reference Šprah, Dernovšek, Wahlbeck and Haaramo43]. Increase in number of relapses increases the risk of development of resistance to the antipsychotic treatment [Reference Kane44]. Therefore, we may hypothesize the association of chronic physical illnesses with resistance to the antipsychotic treatment mediated by relapse rates.
To the best of our knowledge, only a few studies addressed the problem of the association of somatic comorbidities with the outcomes of treatment of psychosis. These studies used different outcome measures, compared to our research, but the prevalence of patients with chronic somatic comorbidities and the main guiding principles were similar to ours. It is important to note the high rates of somatic comorbidities among patients with SSDs, considering higher mortality rates compared to general population [3–Reference Björkenstam, Ljung, Burström, Mittendorfer-Rutz, Hallqvist and Weitoft5], the adverse effects of physical illnesses on patients’ quality of life, lower quality of somatic healthcare and inadequate healthcare strategies for patients with mental illnesses [Reference Björkenstam, Ljung, Burström, Mittendorfer-Rutz, Hallqvist and Weitoft5,Reference Filipčić, Filipčić IŠ, Matić, Lovretić, Ivezić and Bajić Ž12]. The percentage of patients with chronic somatic comorbidities in our study (50%) was exactly the same as what was found in the study by Douzenis et al. [Reference Douzenis, Seretis, Nika, Nikolaidou, Papadopoulou and Rizos45] and similar to that of Chwastiak et al. [Reference Chwastiak, Rosenheck, McEvoy, Keefe, Swartz and Lieberman24] (58%) but, it was higher compared to what was found by Lyketsos et al. [Reference Lyketsos, Dunn, Kaminsky and Breakey23] (21%), possibly due to the fact that Lyketsos et al. studied the general population of psychiatric patients, and patients diagnosed with SSDs made up only 16% of their sample. Also, Lyketsos et al. included inpatients only, whereas our sample comprised both inpatients and outpatients.
The main conclusion of Lyketsos et al. [Reference Lyketsos, Dunn, Kaminsky and Breakey23] was comparable to ours. After the adjustment for illness severity at hospital admission, in their study, somatic comorbidities were significantly associated with higher psychiatric symptom ratings at discharge [Reference Douzenis, Seretis, Nika, Nikolaidou, Papadopoulou and Rizos45]. If the evidences of increase in somatic comorbidities incidence in patients diagnosed with psychosis over the last two decades are correct [3–Reference Björkenstam, Ljung, Burström, Mittendorfer-Rutz, Hallqvist and Weitoft5], the 10-year difference between the Lyketsos et al. and Douzenis et al. studies, as well as the 16-year separation from our study, should be taken into account.
Our findings are apparently contrary to the results of the study conducted by Chwastiak et al. in 2006 [Reference Chwastiak, Rosenheck, McEvoy, Keefe, Swartz and Lieberman24]. They found no significant association between number of chronic physical illnesses and severity of psychiatric symptoms measured by PANSS score in patients diagnosed with schizophrenia. Both of our studies were cross-sectional. However, whereas our key outcome was the number of psychiatric hospitalizations since diagnosis, Chwastiak et al. used the outcome measure that could not reflect the change in psychopathology. They used PANSS score at baseline in a randomized control trial of antipsychotic efficacy in treatment of schizophrenia [Reference Stroup, McEvoy, Swartz, Byerly, Glick and Canive46]. In addition, several important physical illnesses were excluded from their study: myocardial infarction in the previous 6 months, history of or current QTc prolongation, uncompensated congestive heart failure, sustained cardiac arrhythmia, first-degree heart block, and complete left bundle branch block.
Sim et al. in 2006 found the opposite effect of physical illnesses in patients with first episode schizophrenia [Reference Sim, Chan, Chua, Mahendran, Chong and McGorry22]. At 24-month follow-up, patients with physical comorbidity had a significantly greater reduction in total PANSS score: 53% vs. 42% in the group of patients with no somatic illnesses. However, their study population was younger (28 years SD 6.7 years vs. 44 SD 12.9 years in our sample). Any somatic comorbidity was found in 22% vs. 38% in our sample. Finally, their study population was patients in the first episode schizophrenia, whereas the average duration of illness in our sample was 11 (SD 10.1) years. For these reasons, the two studies are not comparable, but the difference in our findings may lead to new hypotheses for future research. It is possible that association of physical illnesses with psychiatric treatment outcomes is different in first and in recurrent episodes.
Our study findings are in line with the findings of Köhler et al. [Reference Köhler, Petersen, Benros, Mors and Gasse25]. In their study, an increased risk of schizophrenia 2-year relapse was significantly associated with somatic comorbidity and concomitant use of nonsteroidal anti-inflammatory drugs (NSAIDs) in painful and inflammatory states, particularly acetylsalicylic acid and diclofenac, except for patients with a prior hospital diagnosis for a musculoskeletal disease, which had a lower risk of relapse.
It is possible that psychiatrists generally pay more attention to mental illnesses and are less sensitive to physical health [Reference Björkenstam, Ljung, Burström, Mittendorfer-Rutz, Hallqvist and Weitoft5,Reference Kopp, Fleischhacker, Stürz, Ruedl, Kumnig and Rumpold47]. Our study indicates that such position would be wrong even from the perspective of treatment of psychosis. Olivares et al.’s systematic literature review from 2013 showed a similar tendency among researchers in schizophrenia. Physical illnesses are rarely analyzed as the possible predictors of schizophrenia relapse [Reference Olivares, Sermon, Hemels and Schreiner48]. Our results indicate that this should not be so, but we should certainly take into account the association of chronic physical illnesses and optimal treatment outcome in patients with SSDs. Unfortunately, our experience in clinical practice over the years indicates that the diagnostic habits in Croatia have not changed very much, which further emphasizes the relevance of our study.
4.1. Limitations of the study
First, our results should not be treated as representative for the total population of Croatian patients diagnosed with SSDs. It is reasonable to assume that the quality of both psychiatric and somatic medicine is better in the center where we enrolled patients than what is the case in provincial hospitals with more sparse resources and less experienced, although not less dedicated, health care professionals. However, this source of bias would probably promote the null hypothesis of no association between number of chronic somatic comorbidities and number of psychiatric rehospitalizations. If so, this limitation would actually strengthen the reliability of our findings. Second, our key independent variable was number of diagnosis of chronic somatic illnesses regardless of the illness severity or type. It is possible that in many somatic illnesses there are moderating effects of severity, illness type, illness duration, and somatic therapy on the association of the number of comorbidities with psychiatric rehospitalizations. Third, we measured only the number of hospital admissions and not the duration nor the reasons for hospitalizations. As Psychiatric Hospital Sveti Ivan, Zagreb is a specialized psychiatric institution, all patients were hospitalized due to psychiatric indication, but we could not differentiate whether the reason for particular rehospitalization was the worsening of the primary psychiatric illness, or because of psychiatric comorbidity. Within the limited time span, number and duration of hospitalizations may even be negatively correlated. Also, no other clinical variables relevant to clinical or treatment outcome were included in this study. However, the number of hospital admissions is a simple and universally understood outcome. As Tom Burns pointed out in 2007, “This understanding may, of course, be more illusory than real; the threshold for admission…may be very different in inner-city London and in a small town in Switzerland” [Reference Burns32]. Fifth, we have not collected the data on the possible confounding effect of antipsychotic doses, although a dosing schedule may be associated with the both: somatic health state and psychosis treatment outcomes. Sixth, no patients’ adherence was analyzed while it may be associated with a larger number of somatic and psychiatric illnesses and a higher rate of psychiatric rehospitalizations. We tried to minimize the consequences of missing data on patients’ adherence by sensitivity analysis where particular antipsychotics were replaced with the binary variable representing long-acting injections vs. oral therapy. Finally, we used a consecutive, and not the random sample from our targeted population what increased the risk of a sampling bias. The strength of our study is that we controlled a relatively large number of possible confounders and we approached the problem from a relatively novel perspective of chronic somatic comorbidities’ impact on the psychosis treatment outcome and not vice versa.
Chronic physical illnesses in patients diagnosed with SSDs are associated with the higher rates of psychiatric rehospitalizations independently of psychiatric comorbidities and other clinical and sociodemographic factors. We need to pay more attention to the treatment of chronic physical illnesses in psychiatric patients not only because of somatic health consequences and patients’ quality of life but also for the effectiveness of psychiatric therapy. The integrative, multidisciplinary approach should be the imperative in clinical practice. “Multimorbidities are indifferent to medical specialties” [Reference Jakovljević and Ostojić40]. Future studies are needed with more rigorously defined and direct psychiatric treatment outcomes and prospective cohort designs that may establish the temporal order of occurrences.
The study was funded by Psychiatric Hospital Sveti Ivan, Zagreb, Croatia.
Disclosure of interest
The authors declare that they have no competing interest.
Authors wish to acknowledge the work of all patients, physicians, and medical nurses who took part in the data collection.