Development and validation of a machine learning model for prediction of type 2 diabetes in patients with mental illness.

BACKGROUND: Type 2 diabetes (T2D) is approximately twice as common among individuals with mental illness compared with the background population, but may be prevented by early intervention on lifestyle, diet, or pharmacologically. Such prevention relies on identification of those at elevated risk (prediction). The aim of this study was to develop and validate a machine learning model for prediction of T2D among patients with mental illness. METHODS: The study was based on routine clinical data from electronic health records from the psychiatric services of the Central Denmark Region. A total of 74,880 patients with 1.59 million psychiatric service contacts were included in the analyses. We created 1343 potential predictors from 51 source variables, covering patient-level information on demographics, diagnoses, pharmacological treatment, and laboratory results. T2D was operationalised as HbA1c ≥48 mmol/mol, fasting plasma glucose ≥7.0 mmol/mol, oral glucose tolerance test ≥11.1 mmol/mol or random plasma glucose ≥11.1 mmol/mol. Two machine learning models (XGBoost and regularised logistic regression) were trained to predict T2D based on 85% of the included contacts. The predictive performance of the best performing model was tested on the remaining 15% of the contacts. RESULTS: The XGBoost model detected patients at high risk 2.7 years before T2D, achieving an area under the receiver operating characteristic curve of 0.84. Of the 996 patients developing T2D in the test set, the model issued at least one positive prediction for 305 (31%). CONCLUSION: A machine learning model can accurately predict development of T2D among patients with mental illness based on routine clinical data from electronic health records. A decision support system based on such a model may inform measures to prevent development of T2D in this high-risk population.

Lexical stability of psychiatric clinical notes from electronic health records over a decade.

OBJECTIVE: Natural language processing (NLP) methods hold promise for improving clinical prediction by utilising information otherwise hidden in the clinical notes of electronic health records. However, clinical practice - as well as the systems and databases in which clinical notes are recorded and stored - change over time. As a consequence, the content of clinical notes may also change over time, which could degrade the performance of prediction models. Despite its importance, the stability of clinical notes over time has rarely been tested. METHODS: The lexical stability of clinical notes from the Psychiatric Services of the Central Denmark Region in the period from January 1, 2011, to November 22, 2021 (a total of 14,811,551 clinical notes describing 129,570 patients) was assessed by quantifying sentence length, readability, syntactic complexity and clinical content. Changepoint detection models were used to estimate potential changes in these metrics. RESULTS: We find lexical stability of the clinical notes over time, with minor deviations during the COVID-19 pandemic. Out of 2988 data points, 17 possible changepoints (corresponding to 0.6%) were detected. The majority of these were related to the discontinuation of a specific note type. CONCLUSION: We find lexical and syntactic stability of clinical notes from psychiatric services over time, which bodes well for the use of NLP for predictive modelling in clinical psychiatry.

Stability of diagnostic coding of psychiatric outpatient visits across the transition from the second to the third version of the Danish National Patient Registry.

OBJECTIVE: In Denmark, data on hospital contacts are reported to the Danish National Patient Registry (DNPR). The ICD-10 main diagnoses from the DNPR are often used as proxies for mental disorders in psychiatric research. With the transition from the second version of the DNPR (DNPR2) to the third (DNPR3) in February-March 2019, the way main diagnoses are coded in relation to outpatient treatment changed substantially. Specifically, in the DNPR2, each outpatient treatment course was labelled with only one main diagnosis. In the DNPR3, however, each visit during an outpatient treatment course is labelled with a main diagnosis. We assessed whether this change led to a break in the diagnostic time-series represented by the DNPR, which would pose a threat to the research relying on this source. METHODS: All main diagnoses from outpatients attending the Psychiatric Services of the Central Denmark Region from 2013 to 2021 (n = 100,501 unique patients) were included in the analyses. The stability of the DNPR diagnostic time-series at the ICD-10 subchapter level was examined by comparing means across the transition from the DNPR2 to the DNPR3. RESULTS: While the proportion of psychiatric outpatients with diagnoses from some ICD-10 subchapters changed statistically significantly from the DNPR2 to the DNPR3, the changes were small in absolute terms (e.g., +0.6% for F2-psychotic disorders and +0.6% for F3-mood disorders). CONCLUSION: The change from the DNPR2 to the DNPR3 is unlikely to pose a substantial threat to the validity of most psychiatric research at the diagnostic subchapter level.

Multiple health complaints in preadolescence and hospital contacts during adolescence: a prospective cohort study.

Multiple health complaints (MHC) is increasing among preadolescents in many countries, but their prognostic effect for individual thriving or societal resource use is scarcely studied. This makes interpreting the significance of this increase challenging. We contribute by examining whether MHC in preadolescence predicts hospital contacts in adolescence by doing a nation-wide population-based cohort-study following preadolescents from the Danish National Birth-Cohort from 2010 to 2018. 96,382 children were invited at age 11. Responses to a modified version of the Health Behaviour in School Children Symptom Checklist (headache, dizziness, stomachache, irritability, feeling nervous, difficulty in getting to sleep and feeling low) was dichotomized into MHC (≥ 2 concurrent symptoms, each with a frequency of at least weekly, yes/no). Hospital contacts were derived from Danish registers from the date of answering the questionnaire to December 31st 2018. Negative binomial regression estimated incidence rate ratios (IRRs) comparing children with MHC to children without. Analyses were further broken down by hospital sector (psychiatric/somatic) and contact type (in-patient/out-patient/emergency room). 47,365 (49.1%) responded. Mean age was 11.2 years, 52% girls. 10.3% of responders reported MHC. For hospital contacts, the unadjusted IRR was 1.74 [95% CI 1.65, 1.83]. Results were robust to adjustment for sociodemographic variables and somatic/psychiatric morbidity diagnosed before baseline, IRR 1.62 [95% CI 1.54-1.71]. In conclusion, MHC in preadolescents are prognostic of hospital contacts. This shows that we cannot ignore MHC, and to prevent potentially unhelpful healthcare use, we must act. Future research should focus on the underlying causes of MHC to understand which changes will be most helpful and thus how to act.

Pre-adolescents with multiple health complaints redeem more prescriptions: A follow-up study in the Danish National Birth Cohort.

Aims: This study aimed to examine the association between multiple health complaints (MHC) in pre-adolescence and prescription redemption in adolescence. Methods: This was a nationwide population-based study based on the Danish National Birth Cohort for an average of 6.9 years (2010-2018). A total of 96,382 children were invited at the age of 11. A modified version of the Health Behaviour in School Children Symptom Checklist was dichotomised into the World Health Organization's definition of MHC (⩾2 complaints, each with a frequency of at least weekly, yes/no). The number of prescriptions was retrieved from Danish registries. Negative binomial regression estimated incidence rate ratios (IRRs) comparing children with MHC to children without. Prescription redemption was further stratified by psychiatric/somatic medication and into subtypes of prescriptions. Results: A total of 47,365 (49.1%) children participated (Mage=11.2 years, 52% girls). MHC were reported by 10.3%. The unadjusted IRR (MHC vs. no MHC) of all types of redemptions was 1.57 (95% confidence interval (CI) 1.49-1.64). Results were robust to adjustment for socio-demographic variables and somatic/psychiatric morbidity at baseline (IRR=1.47; 95% CI 1.40-1.54). Associations were especially strong for psychiatric medication (adjusted IRR=3.88; 95% CI 3.43-4.40) and were modified by neither sex nor maternal education. Conclusions: MHC in pre-adolescents predict prescription redemption. This implies that changes in MHC might be indicative of changes in public health. This requires further study, as the cause of a change in reporting of symptoms might also cause a change in treatment response. The latter determines whether prescriptions are treating ill-being or needlessly medicalising subjective symptoms.

The PSYchiatric clinical outcome prediction (PSYCOP) cohort: leveraging the potential of electronic health records in the treatment of mental disorders.

BACKGROUND: The quality of life and lifespan are greatly reduced among individuals with mental illness. To improve prognosis, the nascent field of precision psychiatry aims to provide personalised predictions for the course of illness and response to treatment. Unfortunately, the results of precision psychiatry studies are rarely externally validated, almost never implemented in clinical practice, and tend to focus on a few selected outcomes. To overcome these challenges, we have established the PSYchiatric Clinical Outcome Prediction (PSYCOP) cohort, which will form the basis for extensive studies in the upcoming years. METHODS: PSYCOP is a retrospective cohort study that includes all patients with at least one contact with the psychiatric services of the Central Denmark Region in the period from January 1, 2011, to October 28, 2020 (n = 119 291). All data from the electronic health records (EHR) are included, spanning diagnoses, information on treatments, clinical notes, discharge summaries, laboratory tests, etc. Based on these data, machine learning methods will be used to make prediction models for a range of clinical outcomes, such as diagnostic shifts, treatment response, medical comorbidity, and premature mortality, with an explicit focus on clinical feasibility and implementation. DISCUSSIONS: We expect that studies based on the PSYCOP cohort will advance the field of precision psychiatry through the use of state-of-the-art machine learning methods on a large and representative data set. Implementation of prediction models in clinical psychiatry will likely improve treatment and, hopefully, increase the quality of life and lifespan of those with mental illness.

A Randomised Trial Examining Cardiovascular Morbidity and All-Cause Mortality 24 years Following General Health Checks: the Ebeltoft Health Promotion Project (EHPP).

INTRODUCTION: Global prevalence of risk factors for cardiovascular disease (CVD) and all-cause mortality is increasing. Treatments are available but can only be implemented if individuals at risk are identified. General health checks have been suggested to facilitate this process. OBJECTIVES: To examine the long-term effect of population-based general health checks on CVD and all-cause mortality. DESIGN AND SETTING: The Ebeltoft Health Promotion Project (EHPP) is a parallel randomised controlled trial in a Danish primary care setting. PARTICIPANTS: The EHPP enrolled individuals registered in the Civil Registration System as (1) inhabitants of Ebeltoft municipality, (2) registered with a general practitioner (GP) participating in the study and (3) aged 30-49 on 1 January 1991. A total of 3464 individuals were randomised as invitees (n=2000) or non-invitees (n=1464). Of the invitees, 493 declined. As an external control group, we included 1 511 498 Danes living outside the municipality of Ebeltoft. INTERVENTIONS: Invitees were offered a general health check and, if test-results were abnormal, recommended a 15-45 min consultation with their GP. Non-invitees in Ebeltoft received a questionnaire at baseline and were offered a general health check at year 5. The external control group, that is, the remaining Danish population, received routine care only. OUTCOME MEASURES: HRs for CVD and all-cause mortality. RESULTS: Every individual randomised was analysed. When comparing invitees to non-invitees within the municipality of Ebeltoft, we found no significant effect of general health checks on CVD (HR=1.11 (0.88; 1.41)) or all-cause mortality (HR=0.93 (0.75; 1.16)). When comparing invitees to the remaining Danish population, we found similar results for CVD (adjusted HR=0.99 (0.86; 1.13)) and all-cause mortality (adjusted HR=0.96 (0.85; 1.09)). CONCLUSION: We found no effect of general health checks offered to the general population on CVD or all-cause mortality. TRIAL REGISTRATION NUMBER: NCT00145782; 2015-57-0002; 62908, 187.