Discussion

To the best of our knowledge, this is the first study to examine physical comorbidity burden trajectories and their association with all-cause mortality among patients with SMI in primary care. Using JLCM, we identified four distinct CCI-defined comorbidity trajectories following the first primary care recording of SMI, each associated with a different level of mortality risk. Several baseline characteristics were significantly associated with trajectory membership, offering potential indicators for early patient stratification and targeted intervention. ^{Reference Foster, Hill, OSullivan and Hancock27}

In our cohort, the persistent-low group comprised 89.9% of patients and maintained a CCI value of 0–1 throughout the 10-year follow-up, with the highest survival rate among all trajectory groups. This aligns with a UK primary care cohort study reporting that most patients with SMI had a low physical multimorbidity burden during the 5 years following diagnosis. ^{Reference Launders, Kirsh, Osborn and Hayes28} Although the CCI does not capture several prevalent conditions in this population, our sensitivity analysis using a more comprehensive measure covering 24 chronic conditions identified a similar proportion (82.3%) in this group, suggesting that the trajectory structure is robust to the choice of multimorbidity measure.

In contrast, a much smaller proportion of patients (5.1%) fell into the persistent-high group, characterised by a consistently elevated CCI of around 4 throughout the follow-up period. This group had a higher mortality risk than the persistent-low group even after adjustment for potential confounders, consistent with evidence that a greater burden of pre-existing physical comorbidities is associated with increased mortality among patients with schizophrenia. ^{Reference Chan, Wong, Yung, Chen and Chang29} Studies in the general older population have also identified a trajectory characterised by persistently high multimorbidity levels. ^{Reference Lee, Joo, Chai and Jun30} In the sensitivity analysis, this group accounted for 8% of the sample, suggesting that a more comprehensive multimorbidity measure captures additional long-term conditions. Patients in this group were older at the index date, and their elevated baseline CCI likely reflects the higher prevalence of chronic conditions associated with ageing rather than factors specific to SMI. ^{Reference St Sauver, Boyd, Grossardt, Bobo, Finney Rutten and Roger31,Reference De Hert, Correll, Bobes, Cetkovich-Bakmas, Cohen and Asai32} Their elevated mortality risk likely reflects the persistent burden of pre-existing comorbidities rather than the development of new conditions during follow-up. The use of antihypertensives, antidiabetics and drugs for obstructive airway disease was notable in this group, consistent with the higher frequency of annual consultations, reflecting greater clinical management needs. A higher proportion of patients had alcohol misuse, which may be explained by the established link between alcohol consumption and various chronic conditions. ^{Reference Stewart and McCambridge33} However, this group had a lower proportion of current smokers and overweight/obese individuals at baseline, which may partly account for the minimal progression of multimorbidity over time.

The remaining two trajectory groups (increase-stable and rapid-increase groups) had similar baseline CCI values of approximately 1, but demonstrated different progression trends and mortality risks, consistent with studies in other populations that have identified distinct upward multimorbidity trajectories. ^{Reference Lee, Joo, Chai and Jun30,Reference Strauss, Jones, Kadam and Jordan34} Compared with the persistent-low group, mortality risk was approximately 4 times as high in the increase-stable group and roughly 11 times as high in the rapid-increase group, which had the highest mortality risk. This suggests that when evaluating the impact of physical comorbidity burden on adverse outcomes, it is important to consider not only the baseline burden, but also its progression over time. Patients in both groups were older at the index date (median 61.8 and 67.5 years, respectively) than those in the persistent-low group (median 41.6 years), consistent with evidence that individuals diagnosed with SMI at an older age tend to have a greater number of physical conditions in the post-SMI period. ^{Reference Wang, Bendayan, Msosa, Pritchard, Roberts and Stewart9} Both groups also had a higher proportion of current smokers and were less likely to be referred to mental health services, which may reflect challenges in timely access to psychiatric care that could, in turn, worsen physical health. ^{Reference Halstead, Sartorius, Every-Palmer, Siddiqi, de Girolamo and Siskind35,Reference Harald, Peltonen and Laatikainen36} Notably, patients in the rapid-increase group were more likely to live in more deprived areas, potentially facing greater barriers to healthcare and healthy lifestyle options. ^{Reference Liu, Daumit, Dua, Aquila, Charlson and Cuijpers37} They also had fewer clinical visits than the increase-stable group despite a similar baseline comorbidity level, suggesting that active engagement with primary care may help stabilise the progression of long-term conditions. Furthermore, the sharp rise in metastatic solid tumours observed in the rapid-increase group likely contributed to its elevated mortality and warrants further investigation in SMI populations. ^{Reference Mani, Deng, Lin, Wang, Hsu and Zaorsky38} However, further investigation is necessary to clarify the effects of mental illness severity and antipsychotic dosage on the rate of physical deterioration.

Health trajectories are strongly age-associated, and we acknowledge that age is likely to have contributed to the observed class separation, particularly given the substantial difference in median age between the persistent-low group and the other high-risk trajectory groups. The identified classes should therefore be interpreted as descriptive, age-related patterns of CCI-defined comorbidity burden, as they manifest after entry into primary care management, rather than as age-independent biological subtypes. Although mortality differences persisted after adjustment for baseline age and other covariates, this does not exclude the possibility that the latent classes largely reflect differences in life-course stages at the index date. Consequently, using the younger persistent-low group as a reference category primarily serves to contrast different levels of observed clinical burden within the cohort. Future studies using larger linked data-sets are needed to examine whether similar trajectory patterns emerge within narrower age subgroups, which would help to further disentangle age-related progression from age-independent comorbidity accumulation.

Our study has several strengths. First, it drew on a large sample of patients with SMI from the THIN primary care database, which has been validated for epidemiological research. Second, routinely collected primary care data enabled an extended follow-up period with sufficient repeated measurements to characterise the evolution of trajectories. Third, the application of JLCM allowed classification to be informed by both CCI trajectories and survival process, thereby reducing bias from informative censoring which can affect conventional group-based trajectory models. However, several limitations should be noted. First, our study relied exclusively on routinely collected primary care data from the THIN database, which introduces several interrelated constraints. The index date may not correspond to the true onset of SMI, as some individuals are likely to have been diagnosed in secondary care before primary care recording; our cohort therefore represents the point of entry into active primary care management. Conditions documented only in secondary care or in GP free text may have been missed, and patients with shorter registration periods may have had comorbidities underrecorded, potentially underestimating the true comorbidity burden. Second, the THIN database does not provide practice-level or regional identifiers because of its anonymisation protocols, preventing the incorporation of area-level random effects; although IMD adjustment partially captures geographical variation, residual confounding by unmeasured area-level factors cannot be excluded. Third, the CCI does not capture several prevalent conditions in this population; however, our sensitivity analysis using a broader 24-condition measure yielded a comparable trajectory structure. Future work could develop a severity-weighted index with broader condition coverage for SMI populations. Fourth, the requirement for at least 1 year of post-index follow-up may have excluded the highest-risk patients who died or transferred out shortly after the index date, potentially underestimating early mortality. Finally, as an observational study, we cannot infer causality. Covariates were measured only at baseline, leaving potential for residual confounding and reverse causation. Missing baseline data, particularly BMI and smoking status, are unlikely to be missing at random, as recording may be influenced by clinical need, which may have introduced residual bias.

In conclusion, our study identified marked heterogeneity in CCI-defined comorbidity trajectories, which were associated with substantially different mortality risks among patients with SMI in primary care. Importantly, the baseline clinical indicators associated with higher-risk trajectories – including older age, smoking and prescriptions for hypertension, diabetes and obstructive airway disease – are routinely collected in primary care. In England, the QOF-mandated annual SMI physical health checks provide a feasible opportunity for risk stratification without introducing additional assessments. Patients likely to follow high-risk trajectories could potentially be prioritised for more intensive physical health monitoring or targeted lifestyle interventions. However, it is important to note that the trajectory groups identified represent latent statistical classifications within this cohort rather than validated clinical phenotypes; future research with larger cohorts, longer follow-up and linked primary and secondary care data is warranted to refine age-specific trajectory patterns and evaluate the clinical effectiveness of stratified care approaches in this vulnerable population.