Researchers have published a detailed protocol for a systematic review intended to identify and compare methodological approaches used in developing clinical prediction models for individuals living with multiple long-term conditions. The study, led by Lauren A Anderson and colleagues, aims to address a significant gap in the current literature where methods designed for singular outcomes have been proven inappropriate for the complex nature of multiple long-term conditions research. By synthesizing existing prognostic model studies, the review seeks to provide a comprehensive overview of the field and inform the development of new models for use in the United Kingdom.
The motivation for this work stems from the rising prevalence of multiple long-term conditions, which are defined as the co-existence of two or more chronic conditions in an individual. These conditions are associated with an increased need for healthcare and a reduced quality of life. Current prevalence estimates range from 0.9% for individuals aged 0 to 19 years to 68.2% in those aged 80 and over, with an average overall prevalence of 14.8%. The odds of developing these conditions are notably increased in deprived communities, prompting funding bodies such as the National Institute of Health and Care Research to prioritize this area of research.
Despite the growing interest, research in this field is limited by varying definitions of multiple long-term conditions and the specific conditions incorporated into studies. To address this inconsistency, an international consensus was developed for the definition and measurement of these conditions in research, comprising 24 specific conditions that should be included. Determining risk factors is essential for improving treatment and intervention strategies, yet current research has largely focused on cross-sectional studies, making it difficult to establish the impact of various determinants on the development of multiple long-term conditions.
Prediction of multiple long-term conditions presents unique challenges due to the complex interactions between co-occurring conditions. Some conditions may have similar aetiology and require similar treatment options, a state known as concordance, while others appear unrelated or require different treatments, known as discordance. Concordant conditions may cluster due to shared risk factors, but discordant conditions may not exhibit such shared factors. Consequently, methodologies must be capable of capturing these diverse outcomes and the interactions between factors that produce condition clusters.
Current treatment approaches are often focused around single conditions, and individuals with multiple long-term conditions frequently experience high levels of polypharmacy, which can lead to harm and the need to manage resulting effects. Furthermore, treatments may need to be considered as confounders during risk prediction to ensure accurate results. Clinical prediction models often consider a singular outcome, but in the case of multiple long-term conditions, there is a need to estimate the risk of co-occurring conditions, which may require a multivariate approach.
Estimating the risk of multiple conditions by multiplying the risks from single-condition models is a common temptation, but this approach operates under the assumption of independence, which is often violated in the context of multiple long-term conditions. There is an absence of clear methodologies to handle this dependence between comorbidities, a problem that remains a significant challenge in the field. This gap may be attributed to the complexity in methodologies, yet bridging it is essential to improving risk estimation for individuals with multiple long-term conditions.
The systematic review will search MEDLINE, Embase, CINAHL, and CENTRAL databases from September 2015 to identify relevant clinical prediction models that predict the development of multiple long-term conditions. The search strategy was developed using the Ingui filter for finding prognostic and diagnostic prediction studies, alongside a specialist librarian, to ensure optimal identification of relevant studies. The filter was adjusted for Embase to ensure suitability, and search terms for multiple long-term conditions were derived from current literature research.
Inclusion criteria for the review specify that studies must involve adults aged 18 and over, with populations that are disease-free or have a single long-term condition at baseline. The studies must report on a clinical prediction model developed to predict the risk of developing more than one new long-term condition. A clinical prediction model is defined as a model incorporating at least two predictors, providing a probability or risk of a future outcome of multiple long-term conditions. Publications from 2015 onwards will be included to focus on contemporary models, ensuring the findings reflect the current state of the field.
Exclusion criteria will remove studies that do not involve the development of a clinical prediction model for more than one long-term condition. Studies only available in abstract form, including conference abstracts and pre-prints, as well as protocols, systematic reviews, and grey literature, will be excluded. Only full-text articles available in English will be included, a restriction driven by the methodological nature of the review, which requires an excellent comprehension of the technicalities of the methods. The number of papers excluded due to language restraints will be reported.
Data extraction will be undertaken independently by two reviewers, with disagreements discussed with a third. The CHARMS checklist has informed the data extraction process, which will include primary items for methodology and model outcomes, as well as secondary items such as study descriptors, population information, and measured outcomes. Model performance measures, including calibration, discrimination, and overall performance statistics, will be extracted, along with internal and external validation data. Clinical usefulness measures, such as net benefit and consideration of impact on health outcomes, will also be recorded.
Risk of bias will be assessed using the PROBAST + AI tool, with two independent reviewers assessing included studies in regard to participants, predictors, outcome, and analysis. Studies will then be rated as low, high, or unclear risk of bias. A narrative synthesis will be conducted to summarize current methodological practice and identify areas for improvement to inform future methodological and model development. Methodologies will be compared and grouped by outcome type, and any restrictions placed upon use, such as the number of outcome conditions, will be analyzed.
The protocol is reported according to the Preferred Reporting Items for Systematic Reviews Protocols (PRISMA-P) guidelines. Ethical approval is not required for this systematic review as it will use published literature only. The findings of the review will be submitted for publication in a peer-reviewed journal. By identifying the strengths and limitations of existing methodologies, the review aims to help better inform current practices and improve model development in this area, ultimately providing better health outcomes for individuals with multiple long-term conditions through improved early detection and condition management.
The review protocol outlines a structured approach to managing the identified studies, beginning with the import of all eligible records into the Covidence platform. Duplicates will be removed before the screening process commences. Titles and abstracts will be evaluated by two independent reviewers against the predefined inclusion and exclusion criteria. Any conflicts arising during this initial screening phase will be resolved through discussion with a third reviewer. Following the abstract screening, a further round of screening will be conducted on the full text of the remaining articles. This stage will also be performed by two independent reviewers, with a third reviewer available to resolve any disagreements. The protocol specifies that the reason for exclusion will be recorded for every paper that does not meet the criteria, ensuring transparency in the selection process.
The data extraction process is designed to capture a wide range of methodological and descriptive details from the included studies. In addition to the primary items concerning methodology and model outcomes, the extraction will include secondary items such as study descriptors, which encompass the title, authors, study dates, publication date, journal, country, setting, and the number of patients involved. Population data will be extracted to document inclusion and exclusion criteria, sample size, age range, baseline condition status, and the number of participants with missing data. The types of long-term conditions measured will also be recorded, providing a detailed overview of the specific health outcomes addressed by the models.
Specific attention will be paid to the methods used in the development of the clinical prediction models. This includes the handling of missing data, the identification of candidate predictors used, and any instances where predictors were combined. The structure of the predictors included in the models will be documented, along with any consideration of proxies. Model performance will be assessed through the extraction of calibration, discrimination, and overall performance statistics. Both internal and external validation data will be captured to evaluate the robustness of the models. Clinical usefulness measures, such as net benefit, will also be extracted, along with any consideration of the impact on health outcomes.
The narrative synthesis will present the main characteristics of the papers, including the methodologies used and key aspects of model development. These findings will be further presented using tables and figures, which will include the risk of bias assessments. Methodologies will be compared and grouped by outcome type, and any restrictions placed upon their use, such as the number of outcome conditions, will be analyzed. Recommendations will be formulated in relation to the application and restrictions of the methods, as well as their ease of use and the associated risk of bias. This approach aims to provide a comprehensive overview of the current methodological landscape.
The protocol acknowledges several strengths and limitations of the planned review. A comprehensive review of all methodologies used in clinical prediction modelling for multiple long-term conditions will be undertaken, with methodologies included regardless of the specific outcome conditions predicted. The protocol is reported according to the Preferred Reporting Items for Systematic Reviews Protocol (PRISMA-P) guidelines. However, the review is subject to the limitation that only English-language papers will be included. This restriction is due to the methodological nature of the review, which requires an excellent comprehension of the technicalities of the methods. The number of papers excluded due to language restraints will be reported to provide transparency regarding this limitation.
Subject of Research: Protocol for a systematic review of clinical prediction models for multiple long-term conditions
Article Title: A comparison of methodological approaches to developing clinical prediction models for individuals living with multiple long-term conditions: a protocol for a systematic review
Article References: Anderson, L. A., Ensor, J., Gillies, C. L., Lock, S. T., Khunti, K., & Gray, L. J. (2026). A comparison of methodological approaches to developing clinical prediction models for individuals living with multiple long-term conditions: a protocol for a systematic review. Diagnostic and Prognostic Research, 10(1), Article 6. https://doi.org/10.1186/s41512-026-00221-2
Image Credits: AI Generated
DOI: 10.1186/s41512-026-00221-2
Keywords: multiple long-term conditions, clinical prediction models, systematic review, methodology, risk prediction, multimorbidity, health research, prognostic models, comparison, methodological, approaches, developing
News Source: Ophelia Keating. (November 7, 2026). Protocol outlines review of MLTC prediction models. Scienmag.
