Prediction models using artificial intelligence and longitudinal data from electronic health records: a systematic methodological review.

OBJECTIVE: To describe and appraise the use of artificial intelligence (AI) techniques that can cope with longitudinal data from electronic health records (EHRs) to predict health-related outcomes. METHODS: This review included studies in any language that: EHR was at least one of the data sources, collected longitudinal data, used an AI technique capable of handling longitudinal data, and predicted any health-related outcomes. We searched MEDLINE, Scopus, Web of Science, and IEEE Xplorer from inception to January 3, 2022. Information on the dataset, prediction task, data preprocessing, feature selection, method, validation, performance, and implementation was extracted and summarized using descriptive statistics. Risk of bias and completeness of reporting were assessed using a short form of PROBAST and TRIPOD, respectively. RESULTS: Eighty-one studies were included. Follow-up time and number of registers per patient varied greatly, and most predicted disease development or next event based on diagnoses and drug treatments. Architectures generally were based on Recurrent Neural Networks-like layers, though in recent years combining different layers or transformers has become more popular. About half of the included studies performed hyperparameter tuning and used attention mechanisms. Most performed a single train-test partition and could not correctly assess the variability of the model's performance. Reporting quality was poor, and a third of the studies were at high risk of bias. CONCLUSIONS: AI models are increasingly using longitudinal data. However, the heterogeneity in reporting methodology and results, and the lack of public EHR datasets and code sharing, complicate the possibility of replication. REGISTRATION: PROSPERO database (CRD42022331388).

Contribution of Frailty to Multimorbidity Patterns and Trajectories: Longitudinal Dynamic Cohort Study of Aging People.

BACKGROUND: Multimorbidity and frailty are characteristics of aging that need individualized evaluation, and there is a 2-way causal relationship between them. Thus, considering frailty in analyses of multimorbidity is important for tailoring social and health care to the specific needs of older people. OBJECTIVE: This study aimed to assess how the inclusion of frailty contributes to identifying and characterizing multimorbidity patterns in people aged 65 years or older. METHODS: Longitudinal data were drawn from electronic health records through the SIDIAP (Sistema d'Informació pel Desenvolupament de la Investigació a l'Atenció Primària) primary care database for the population aged 65 years or older from 2010 to 2019 in Catalonia, Spain. Frailty and multimorbidity were measured annually using validated tools (eFRAGICAP, a cumulative deficit model; and Swedish National Study of Aging and Care in Kungsholmen [SNAC-K], respectively). Two sets of 11 multimorbidity patterns were obtained using fuzzy c-means. Both considered the chronic conditions of the participants. In addition, one set included age, and the other included frailty. Cox models were used to test their associations with death, nursing home admission, and home care need. Trajectories were defined as the evolution of the patterns over the follow-up period. RESULTS: The study included 1,456,052 unique participants (mean follow-up of 7.0 years). Most patterns were similar in both sets in terms of the most prevalent conditions. However, the patterns that considered frailty were better for identifying the population whose main conditions imposed limitations on daily life, with a higher prevalence of frail individuals in patterns like chronic ulcers &peripheral vascular. This set also included a dementia-specific pattern and showed a better fit with the risk of nursing home admission and home care need. On the other hand, the risk of death had a better fit with the set of patterns that did not include frailty. The change in patterns when considering frailty also led to a change in trajectories. On average, participants were in 1.8 patterns during their follow-up, while 45.1% (656,778/1,456,052) remained in the same pattern. CONCLUSIONS: Our results suggest that frailty should be considered in addition to chronic diseases when studying multimorbidity patterns in older adults. Multimorbidity patterns and trajectories can help to identify patients with specific needs. The patterns that considered frailty were better for identifying the risk of certain age-related outcomes, such as nursing home admission or home care need, while those considering age were better for identifying the risk of death. Clinical and social intervention guidelines and resource planning can be tailored based on the prevalence of these patterns and trajectories.

Dynamics of multimorbidity and frailty, and their contribution to mortality, nursing home and home care need: A primary care cohort of 1 456 052 ageing people.

Background: Prevalence of both multimorbidity and frailty increases with age, but more evidence is needed to elucidate their relationship and their association with other health-related outcomes. We analysed the dynamics of both conditions as people age and calculate the associated risk of death, nursing home admission, and need for home care. Methods: Data were drawn from the primary care electronic health records of a longitudinal cohort of people aged 65 or older in Catalonia in 2010-2019. Frailty and multimorbidity were measured using validated instruments (eFRAGICAP, a cumulative deficit model; and SNAC-K, respectively), and their longitudinal evolution was described. Cox regression models accounted for the competing risk of death and adjusted by sex, socioeconomical status, and time-varying age, alcohol and smoking. Findings: We included 1 456 052 patients. Prevalence of multimorbidity was consistently high regardless of age, while frailty almost quadrupled from 65 to 99 years. Frailty worsened and also changed with age: up to 84 years, it was more related to concurrent diseases, and afterwards, to frailty-related deficits. While concurrent diseases contributed more to mortality, frailty-related deficits increased the risk of institutionalisation and the need for home care. Interpretation: The nature of people's multimorbidity and frailty vary with age, as does their impact on health status. People become frailer as they age, and their frailty is more characterised by disability and other symptoms than by diseases. Mortality is most associated with the number of comorbidities, whereas frailty-related deficits are associated with needing specialised care. Funding: Instituto de Salud Carlos III through PI19/00535, and the PFIS Grant FI20/00040 (Co-funded by European Regional Development Fund/European Social Fund).

Autoencoders for health improvement by compressing the set of patient features.

A challenge to solve when analyzing multimorbidity patterns in elderly people is the management of a high number of characteristics associated with each patient. The main variables to study multimorbidity are diseases, however other variables should be considered to better classify the people included in each pattern. Age, sex, social class and medication are frequently used in the typing of each multimorbidity pattern. Subsequently the cardinality of the set of features that characterize a patient is very high and normally, the set is compressed to obtain a patient vector of new variables whose dimension is noticeably smaller than that of the initial set. To minimize the loss of information by compression, traditionally Principal Component Analysis (PCA) based projection techniques have been used, which although they are generally a good option, the projection is linear, which somehow reduces its flexibility and limits the performance. As an alternative to the PCA based techniques, in this paper, it is proposed to use autoencoders, and it is shown the improvement in the obtained multimorbidity patterns from the compressed database, when the registered data on about a million patients (5 years' follow-up) are processed. This work demonstrates that autoencoders retain a larger amount of information in each pattern and results are more consistent with clinical experience than other approaches frequently found in the literature.Clinical relevance- From an epidemiological perspective, the contribution is relevant, since it allows for a more precise analysis of multimorbidity patterns, leading to better approaches to patient health strategies.

Five-year trajectories of multimorbidity patterns in an elderly Mediterranean population using Hidden Markov Models.

This study aimed to analyse the trajectories and mortality of multimorbidity patterns in patients aged 65 to 99 years in Catalonia (Spain). Five year (2012-2016) data of 916,619 participants from a primary care, population-based electronic health record database (Information System for Research in Primary Care, SIDIAP) were included in this retrospective cohort study. Individual longitudinal trajectories were modelled with a Hidden Markov Model across multimorbidity patterns. We computed the mortality hazard using Cox regression models to estimate survival in multimorbidity patterns. Ten multimorbidity patterns were originally identified and two more states (death and drop-outs) were subsequently added. At baseline, the most frequent cluster was the Non-Specific Pattern (42%), and the least frequent the Multisystem Pattern (1.6%). Most participants stayed in the same cluster over the 5 year follow-up period, from 92.1% in the Nervous, Musculoskeletal pattern to 59.2% in the Cardio-Circulatory and Renal pattern. The highest mortality rates were observed for patterns that included cardio-circulatory diseases: Cardio-Circulatory and Renal (37.1%); Nervous, Digestive and Circulatory (31.8%); and Cardio-Circulatory, Mental, Respiratory and Genitourinary (28.8%). This study demonstrates the feasibility of characterizing multimorbidity patterns along time. Multimorbidity trajectories were generally stable, although changes in specific multimorbidity patterns were observed. The Hidden Markov Model is useful for modelling transitions across multimorbidity patterns and mortality risk. Our findings suggest that health interventions targeting specific multimorbidity patterns may reduce mortality in patients with multimorbidity.

Soft clustering using real-world data for the identification of multimorbidity patterns in an elderly population: cross-sectional study in a Mediterranean population.

OBJECTIVES: The aim of this study was to identify, with soft clustering methods, multimorbidity patterns in the electronic health records of a population ≥65 years, and to analyse such patterns in accordance with the different prevalence cut-off points applied. Fuzzy cluster analysis allows individuals to be linked simultaneously to multiple clusters and is more consistent with clinical experience than other approaches frequently found in the literature. DESIGN: A cross-sectional study was conducted based on data from electronic health records. SETTING: 284 primary healthcare centres in Catalonia, Spain (2012). PARTICIPANTS: 916 619 eligible individuals were included (women: 57.7%). PRIMARY AND SECONDARY OUTCOME MEASURES: We extracted data on demographics, International Classification of Diseases version 10 chronic diagnoses, prescribed drugs and socioeconomic status for patients aged ≥65. Following principal component analysis of categorical and continuous variables for dimensionality reduction, machine learning techniques were applied for the identification of disease clusters in a fuzzy c-means analysis. Sensitivity analyses, with different prevalence cut-off points for chronic diseases, were also conducted. Solutions were evaluated from clinical consistency and significance criteria. RESULTS: Multimorbidity was present in 93.1%. Eight clusters were identified with a varying number of disease values: nervous and digestive; respiratory, circulatory and nervous; circulatory and digestive; mental, nervous and digestive, female dominant; mental, digestive and blood, female oldest-old dominant; nervous, musculoskeletal and circulatory, female dominant; genitourinary, mental and musculoskeletal, male dominant; and non-specified, youngest-old dominant. Nuclear diseases were identified for each cluster independently of the prevalence cut-off point considered. CONCLUSIONS: Multimorbidity patterns were obtained using fuzzy c-means cluster analysis. They are clinically meaningful clusters which support the development of tailored approaches to multimorbidity management and further research.

Counting malaria parasites with a two-stage EM based algorithm using crowsourced data.

Malaria eradication of the worldwide is currently one of the main WHO's global goals. In this work, we focus on the use of human-machine interaction strategies for low-cost fast reliable malaria diagnostic based on a crowdsourced approach. The addressed technical problem consists in detecting spots in images even under very harsh conditions when positive objects are very similar to some artifacts. The clicks or tags delivered by several annotators labeling an image are modeled as a robust finite mixture, and techniques based on the Expectation-Maximization (EM) algorithm are proposed for accurately counting malaria parasites on thick blood smears obtained by microscopic Giemsa-stained techniques. This approach outperforms other traditional methods as it is shown through experimentation with real data.