Demographic and health profiles of people with severe mental illness in general practice in Australia: a cross-sectional study

Keywords: bipolar disorder, chronic diseases, comorbidity, general practice, health inequality, preventive health, schizophrenia, severe mental illness.

Severe mental illness (SMI; i.e. psychotic illness, primarily schizophrenia or bipolar affective disorder, severe depression) presents a significant global health equity challenge (Gronholm et al. 2021). It affects about 4% of the population worldwide (Gronholm et al. 2021) and 3% of Australians each year (Whiteford et al. 2017). People with SMI die approximately 20 years earlier than those in the general population (Gronholm et al. 2021). Most deaths are due to preventable physical health conditions (e.g. cardiovascular disease, respiratory illnesses, diabetes, and cancer) (Firth et al. 2019; Gronholm et al. 2021). Higher levels of morbidity and mortality are due to numerous factors including modifiable risk factors, psychiatric medicines, healthcare and socio-economic factors (Morgan et al. 2017; Firth et al. 2019). People with SMI are more likely to smoke, excessively drink alcohol, eat poor diets and be physically inactive than the general population (Firth et al. 2019). They are also less likely to receive preventive care (e.g. weight and blood pressure monitoring; assessment and treatment for high cholesterol) than other patients (Firth et al. 2019).

General practice has an important role in addressing this health inequity. General practitioners (GPs) are the main source of preventive and other health care, with 83% of Australians seeing a GP each year (Australian Institute of Health and Welfare 2020). The Australian Survey of High Impact Psychosis (SHIP) identified that 88% of adults who screened positive for psychosis had seen a GP the previous year (Raudino et al. 2014).

There is little research on the prevalence and characteristics of people with SMI attending general practices in Australia and their health risk factors and comorbidities. Surveys from different samples of people with different mental illness profiles provide different demographic and health trends. For example, general practice patients tend to be better functioning than patients under the care of mental health services (Carr et al. 2002). The impacts of mild-to-moderate mental health conditions are not as extensive as SMI (Firth et al. 2019). Consequently, data from general practice patients with SMI are needed to inform general practice policy and practice about this group. How much of the GP workload includes people with SMI? How do these patients differ from other patients? What does general practice data tell us about the health risk indicators and comorbidities of patients with SMI?

This study aimed to estimate the prevalence of the two main types of SMI (schizophrenia and bipolar disorder) in a nationwide sample of patients regularly attending Australian general practices. The focus on these two conditions was related to their high contribution to the burden of disease (Ferrari et al. 2016; Charlson et al. 2018), as well as the difficulties in identifying patients with severe levels of other severe mental illnesses within Electronic Health Records (EHR) data. The occurrence of common comorbidities and health risk factors was assessed to inform better practices for prevention and management of the physical health of this population.

This was a cross-sectional observational study using MedicineInsight data from a national program managed by National Prescribing Service (NPS) MedicineWise (Busingye et al. 2019; NPS MedicineWise 2020). It contains longitudinal de-identified data extracted from EHRs from Australian general practices, including data equivalent to 13% of patients attending general practices in 2018–19 (NPS MedicineWise 2020). The profile of patients in the MedicineInsight dataset was similar to the profile of people who visited a GP during 2018–19 in terms of age, sex and socioeconomic status (NPS MedicineWise 2020). This study used a random sample of data from 25% of patients provided under contract to the UNSW School of Public Health (N = 3 473 336).

The dataset included ‘condition flags’ created by NPS MedicineWise that indicated if a condition was reported at any time in the MedicineInsight database for each patient (NPS MedicineWise 2019). These algorithms have shown good to excellent accuracy in identifying patients with five prevalent health conditions (sensitivity ≥85%) (Havard et al. 2021).

Adult patients (18–110 years) who were not deceased and regular patients of 565 practices as of 31 December 2018 were included in this study. Regular patients were those with at least three consultations in the previous 2 years in a single general practice (Royal Australian College of General Practitioners 2017). This criterion improved the quality of data ascertainment, increasing the opportunity for clinicians to record relevant health conditions, prescriptions, and tests (Busingye et al. 2019).

Age was calculated at 30 December 2018, based on each patient’s year of birth. Patients with SMI were those with condition flags for schizophrenia or bipolar disorders (SBD) (Supplementary Table S1). Major depressive disorder was not included because the severity of depression was not reliably recorded in the dataset (NPS MedicineWise 2020).

Patients’ characteristics included: (i) socio-demographic profile (age, sex, Indigenous status, government benefits); (ii) health risk factors (smoking status, substance use, dyslipidaemias, hypercholesterolemia, hypertriglyceridemia and hypertension); and (iii) chronic health conditions (cardiovascular diseases, chronic kidney diseases, diabetes mellitus type 2 (T2DM), chronic respiratory conditions, cancer, chronic liver diseases) (Supplementary Table S1). Body mass index was not included due to inconsistent recording of weight for two-thirds of patients (NPS MedicineWise 2020).

Smoking status was based upon the most recent recording of smoking status. Variables for other risk factors and the health conditions used in this study were NPS ‘condition flags’. The condition flag for substance abuse was broadly defined to include, for example, smoking; alcohol use, abuse and dependence; illicit drug use, abuse and dependence; and prescription drug abuse. Given this broad definition, the term ‘substance use’ is used in this paper.

Practice characteristics included rurality and the Index of Relative Socio-Economic Advantage and Disadvantage (IRSAD) quintiles (Australian Bureau of Statistics 2018).

The prevalence of SBD was calculated by dividing the number of patients with a condition flag for SBD in the EHRs by the total number of included patients. Results were presented as a percentage with 95% confidence intervals (CIs) adjusted for clustering.

Descriptive statistics were calculated to assess the distribution of socio-demographic characteristics, health risk factors, and comorbidities in the study population stratified by study group (patients with and without SBD). Results are presented as percentages and associated 95% CIs for all categorical variables. Differences between patients with and without SBD and practice characteristics were assessed using the Chi-squared test. The 95% CIs were calculated using robust variance estimation to account for clustering by practice.

The association between a record of SBD and the probability of having risk factors or comorbidities was evaluated by multilevel models adjusted for age and sex (model 1). As the likelihood of recorded SBD related to practices’ IRSAD and remoteness of the area, multilevel models were further adjusted for IRSAD and remoteness (model 2). The practice ID was the grouping variable for the multilevel models. Model 2 presented a better fit than model 1 (assessed by the lower Akaikes’ Information Criterion), so was the final model in this study.

All analyses were performed using SAS software version 9.4 (SAS Institute Inc., Cary, NC, USA). ‘PROC SURVEY’ was used to estimate CIs by robust variance; ‘PROC GLIMMIX’ with the LAPLACE method was used for the multilevel models.

Of 618 849 patients who met the study criteria, 11 813 (1.91%, 95% CI = 1.88%–1.94%) had a record of SBD.

Patients with a record of SBD were more often males (48% vs 43%) than patients with no record of SBD. They were also more likely to be aged 30–59 years (59% vs 47%), to be recorded as Aboriginal or Torres Strait Islanders (6% vs 2%), have a healthcare card (67% vs 31%) and to have attended practices from the most disadvantaged areas (Table 1). Patient with recorded SBD had a lower proportion of missing data for recording of Indigenous status (17% vs 21%) and government benefits (9% vs 17%).

Table 1.  Sociodemographic characteristics of patients regularly attending general practices in Australia according to records of schizophrenia and bipolar disorders (SBD) diagnosis, 2018 (N = 618 849).

SBD patients had significantly higher prevalence rates than patients without SBD for almost all the health risk factors: smoking (39% vs 13%), substance use (16% vs 2.5%), dyslipidaemia (22% vs 18%), hypercholesterolemia (13% vs 11%), and hypertriglyceridemia (0.3% vs 0.1%). Hypertension was significantly more prevalent among patients without SBD (22% vs 20%) (Table 2).

Table 2.  Prevalence of health behaviour risk factors and health conditions among patients with records of schizophrenia and bipolar (SBD) and patients without SBD in general practices in Australia (N = 618 849).

After adjusting for confounders, the probability of all health risk factors was higher among SBD patients than patients without SBD (adjusted odds ratio (aOR) 1.1–5.9). The highest probabilities were for substance use (aOR = 5.9), smoking (aOR = 3.8) and hypertriglyceridemia (aOR = 3.3). The probabilities of recorded dyslipidaemias, hypercholesterolemia, and hypertension were from 10% to 60% higher among SBD patients than patients without SBD (Table 3).

Table 3.  Probability of health risk factors among patients with records of schizophrenia and bipolar disorders (SBD) compared to patients without SBD (N = 618 849).

SBD patients had higher recorded rates of most comorbidities evaluated than their counterparts, including chronic kidney disease (1.7% vs 1.4%), T2DM (9.9% vs 6.0%), chronic respiratory conditions (25% vs 16%) and chronic liver disease (0.7% vs 0.3%). However, unadjusted rates of cardiovascular disease and cancer were higher among patients without SBD (10% vs 9% and 15% vs 13%, respectively; Table 2).

Adjusted results revealed SBD patients were more likely to be diagnosed with all comorbidities assessed. The probability of having a recorded chronic liver disease diagnosis was over 3-fold higher among SBD patients (aOR = 3.3); T2DM was approximately 2-fold higher (aOR = 2.2). The probabilities of having recorded cardiovascular disease, chronic obstructive pulmonary disease, chronic kidney disease, or cancer were 10–70% higher among SBD patients than patients without SBD (Table 3).

This study estimated the prevalence of recorded SBD as 2% in a population regularly attending Australian general practices. SBD patients were more likely to be male, less likely to be in the youngest or oldest age categories, and more likely to attend practices from the most disadvantaged areas compared to patients without SBD. They were also more likely to be recorded as Aboriginal or Torres Strait Islanders and in receipt of government benefits. Moreover, SBD patients had a higher prevalence of recorded health risk factors and chronic health conditions than non-SBD patients after adjusting for confounders.

The prevalence of 2% was lower than other estimates of SMI, as it only included those with a record of SBD. It also did not include patients with depression or anxiety that might warrant a classification of SMI. The data were insufficient for identifying such patients. As such, the prevalence of 2% seems consistent with Whiteford et al.’s estimate of SMI being 3.3% of the Australian population (Morgan et al. 2017). Although this might seem a small proportion of patients, they contribute substantially to the burden of disease (Ferrari et al. 2016; Charlson et al. 2018).

There were different age distributions of patients with and without SBD. Patients from the SBD group were less likely to be in the youngest age group, possibly related to SBD being not yet diagnosed (Solmi et al. 2022). They were also less likely to be above the age of 60 years, which could reflect their shorter life expectancy compared with other patients (Firth et al. 2019).

The high levels of indicators of socio-economic disadvantage were consistent with other Australian research about people with psychotic disorders; 67% or our sample had a pension or concession card recorded compared to 85% of respondents in the Australian Survey of High Impact Psychosis (SHIP) (Morgan et al. 2017). The lower proportion of patients with a healthcare card in our sample could be due to incomplete records and a selection effect. Socioeconomic disadvantage among people with SBD is well known (Morgan et al. 2017), and our study provides evidence for this among those who access a GP.

A higher proportion of SBD patients were recorded as having an Aboriginal or Torres Strait Islander background than non-SBD patients. This could be related to higher rates of schizophrenia (4.8% vs 1.5%) and bipolar affective disorder (3% vs 1.7%) among Indigenous Australians compared with non-Indigenous Australians (Australian Institute of Health and Welfare 2016). There has been a lack of community population prevalence data for psychotic disorders in Indigenous Australians populations (Black et al. 2015). This study adds to the need for urgent action to address the double jeopardy of health inequities experienced by people with SMI and Aboriginal and Torres Strait Islander peoples more widely.

The higher rates of health risks and comorbidities for patients with SBD compared with other patients (without SBD) were consistent with population surveys of people with SMI (Morgan et al. 2017; Firth et al. 2019) and the NPS analysis of general practice patients with long-term mental illness (LTMI) (NPS MedicineWise 2020). The SHIP identified higher rates of substance use, smoking, hypertension, dyslipidemia, hyperglycaemia, elevated triglyceride levels, and diabetes, compared with general population data (Morgan et al. 2012, 2017). The NPS estimated patients with LTMI had significantly higher adjusted odds for smoking, dyslipidaemia, cardiovascular diseases and diabetes. Direct comparisons between studies are difficult because each study included different mental health conditions and adjusted for a different set of potential confounders. To illustrate this point, our adjusted odds ratio for smoking was 3.8; SHIP’s was 2.3 and the NPS’s was 2.5 (Morgan et al. 2017).

The prevalence estimates of risk factors and comorbidities tended to be lower in the present study than in population surveys. The SHIP reported higher rates than the present study for smoking (66% vs 39%), substance use (about half vs 16%), heart or circulatory disease (27% vs 10%), hypertension (39% vs 20%), diabetes (35% vs 10%), and asthma (30% vs 25% including other chronic respiratory conditions) (Morgan et al. 2012). Given existing evidence on the incomplete recording of risk factors in EHRs in general practice (Harris et al. 2017; NPS MedicineWise 2020), this is probably in part due to incomplete capture of those conditions in the EHRs compared to surveys. Another reason could be the healthier profile of people regularly attending general practices (Carr et al. 2002).

The different pattern of results between raw prevalence data and adjusted odds for hypertension, cardiovascular disease (CVD) and cancer highlights the need for caution when comparing unadjusted prevalence figures. For example, this study and the SHIPS study identified that the prevalence of cancer was lower for each sample of SMI than the comparison data; however, after adjustment, our study identified that the odds of cancer were in fact greater for the SBD group.

Although the higher rates of recorded risk factors and comorbidities for SBD patients suggested that these conditions were screened and monitored, it is still possible that some patients with SMI missed screening for cardiometabolic risk. However, Morgan et al. (2017) identified a decline in the proportion of people with psychosis who had received a physical health examination between 1998 and 2010 (from 80% to 66%) or a fasting blood test (from 83% to 65%) in the previous 12 months (Morgan et al. 2017). Similarly, the NPS identified that patients with LTMI and established CVD were less likely to be prescribed recommended therapy for cardiovascular prevention than patients without LTMI and existing CVD (NPS MedicineWise 2020).

The higher rates of risk factors and comorbidities suggests that more preventive management is needed, particularly given evidence of excess premature mortality among people with SMI (Firth et al. 2019). Recommendations for service delivery have been published by the World Health Organization (WHO) and others (World Health Organization 2018; Firth et al. 2019). Programs have been developed to improve access to general practice by people with SMI (Cameron et al. 2017); however, access to preventive care remains an issue for people with SMI across health systems and urgently needs to be addressed.

Even if access and screening were improved, are there interventions GPs can offer to their patients with SMI? Factors such as cognitive deficits and the effects of medication pose challenges for lifestyle interventions, so adaptation of existing interventions are required (Firth et al. 2019). There is some promising research on interventions to address smoking (Peckham et al. 2017), diet (Teasdale et al. 2017) and physical activity (Rosenbaum et al. 2014) in this patient group. Beyond this, better control of comorbidities such as diabetes and hypertension are needed, including better integration of care across different services and providers (Firth et al. 2019).

A limitation of this study was the completeness of clinical records in the EHRs. MedicineInsight does not include information on all of the factors that contribute to the health gap for people with severe mental illness (e.g. housing security). We have focused on the data available and limited our discussion to this. Although 39% of SBD patients were recorded as smokers, this was likely an underestimate, given overall levels of recording of smoking by GPs (Selak et al. 2006). This study might not have captured all patients with SBD, because they had not been diagnosed or their diagnosis was not recorded in the EHRs. Further, condition flags could not distinguish between the absence of a condition and missing data. Consequently, the degree to which group differences were due to differences in rates of health conditions or differences in recording is unknown. The differences in missing data between SBD and other patients for Indigenous status (17% vs 21%) and government benefits (9% vs 17%) show how differences in prevalence rates can be influenced by rates of missing data between the two groups. This is an important distinction that warrants further research. General practice EHRs do not systematically record the reasons that screening is not performed. In addition, we were not able to assess patterns of psychotropics prescribing due to complexities in retrieving this information from the MedicineInsight dataset. Finally, this study included only patients regularly attending general practices, and their health profiles might not be generalisable to patients that do not visit their GP regularly, including people with unstable housing.

This study estimated the prevalence of SBD in Australian general practice and the likelihood of comorbidities and health risk factors among this population compared to patients without SBD. Preventive care in general practice for this disadvantaged population is important to reduce the gap in health outcomes. Further research is required to explore the quality of health care provided (preventive and therapeutic) by GPs for patients with SBD to better understand how primary care can contribute to reducing disparities.

The data that support this study were obtained from the NPS MedicineWise by licence. Data can only be obtained directly from the NPS MedicineWise.

The authors declare no conflicts of interest.

This research was supported by the National Health and Medical Research Council (NHMRC) Centre of Research Excellence in Medicines Intelligence (ID: 1196900). The Centre for Primary Health Care and Equity, UNSW Sydney, funded access to the MedicineInsight data via the School of Public Health, UNSW Sydney.

Supplementary material is available online.