Machine learning-derived asthma and allergy trajectories in children: a systematic review and meta-analysis

Abstract

Introduction: Numerous studies have characterised trajectories of asthma and allergy in children using machine learning, but with different techniques and mixed findings. The present work aimed to summarise the evidence and critically appraise the methodology.

Methods: 10 databases were searched. Screening, data extraction and quality assessment were performed in pairs. Trajectory characteristics were tabulated and visualised. Associated risk factor and outcome estimates were pooled using a random-effects meta-analysis.

Results: 89 studies were included. Early-onset (infancy) persistent, mid-onset (∼2-5 years) persistent, early-onset early-resolving (within ∼2 years) and early-onset mid-resolving (by ∼3-6 years) wheezing and eczema, respectively, were the most commonly identified disease trajectories. Intermediate/transient trajectories were rare. Male sex was associated with a higher risk of most wheezing trajectories and possibly with early-resolving eczema, while being slightly protective against mid-onset persistent eczema. Parental disease/genetic markers were associated with persistent trajectories of wheezing and eczema, respectively. Prenatal (and less so postnatal) tobacco smoke exposure was associated with most wheezing trajectories, as were lower respiratory tract infections in infancy (particularly with the early-onset resolving patterns). Most studies (69%) were of low methodological quality (particularly in modelling approaches and reporting). Few studies investigated allergic multimorbidity, allergic rhinitis and food allergy.

Conclusions: Childhood asthma/wheezing and eczema can be characterised by a few relatively consistent trajectories, with some actionable risk factors such as pre-/postnatal smoke exposure. Improved computational methodology is warranted to better assess generalisability and elucidate the validity of intermediate/transient trajectories. Likewise, allergic multimorbidity and trajectories of allergic rhinitis and food allergy need to be further elucidated.

FIGURE 1

Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow chart. The “Referenced reports from year ≤2012” represents the papers published in 2012 or earlier in which trajectories were derived and which were referenced in secondary analyses (published within the defined timeframe (2013–2023)) included in this work. AI: artificial intelligence; CINAHL: Cumulative Index to Nursing and Allied Health Literature; WHO: World Health Organization.

FIGURE 2

Early-onset and mid-onset persistent wheezing. Each line represents a trajectory from a study. When multiple trajectories are included from the same study, different line types (e.g. solid or dashed) distinguish between them. Line colour indicates the specific study source for each trajectory. Alongside each study, a percentage shows the proportion of the study's total sample represented by that trajectory. The points along each line denote the time-points at which an assessment of the trajectory-defining variable(s) was done. The y-axis denotes the prevalence/probability, while the x-axis indicates the subject age. a) Early-onset persistent wheezing in studies [21, 34, 36, 44, 46, 59, 67, 68, 81, 87, 91] which presented trajectories by probability. b) Early-onset persistent wheezing in studies [25, 32, 37, 79, 86] which presented trajectories by prevalence. c) Mid-onset persistent wheezing in studies [30, 34, 36, 44, 46, 59, 81] which presented trajectories by probability. d) Mid-onset persistent wheezing in studies [23, 32, 37, 55, 86] which presented trajectories by prevalence.

FIGURE 3

Early-onset mid-resolving and early-onset early-resolving wheezing. Each line represents a trajectory from a study. When multiple trajectories are included from the same study, different line types (e.g. solid, dotted or dashed) distinguish between them. Line colour indicates the specific study source for each trajectory. Alongside each study, a percentage shows the proportion of the study's total sample represented by that trajectory. The points along each line denote the time-points at which an assessment of the trajectory-defining variable(s) was done. The y-axis denotes the prevalence/probability, while the x-axis indicates the subject age. a) Early-onset mid-resolving wheezing in studies [21, 30, 34, 44, 46, 67, 81, 87, 91] which presented trajectories by probability. b) Early-onset mid-resolving wheezing in studies [25, 32, 37, 55, 79, 86] which presented trajectories by prevalence. c) Early-onset early-resolving wheezing in studies [30, 36, 59, 68] which presented trajectories by probability.

FIGURE 4

Early-onset and mid-onset persistent eczema. Each line represents a trajectory from a study. When multiple trajectories are included from the same study, different line types (e.g. solid or dashed) distinguish between them. Line colour indicates the specific study source for each trajectory. Alongside each study, a percentage shows the proportion of the study's total sample represented by that trajectory. The points along each line denote the time-points at which an assessment of the trajectory-defining variable(s) was done. The y-axis denotes the prevalence/probability, while the x-axis indicates the subject age. a) Early-onset persistent eczema in studies [24, 48, 61, 74, 83, 90, 93] which presented trajectories by probability. b) Early-onset persistent eczema in studies [65, 71] which presented trajectories by prevalence. c) Mid-onset persistent eczema in studies [24, 61, 74, 83, 90] which presented trajectories by probability.

FIGURE 5

Early-onset mid-resolving and early-onset early-resolving eczema. Each line represents a trajectory from a study. When multiple trajectories are included from the same study, different line types (e.g. solid or dashed) distinguish between them. Line colour indicates the specific study source for each trajectory. Alongside each study, a percentage shows the proportion of the study's total sample represented by that trajectory. The points along each line denote the time-points at which an assessment of the trajectory-defining variable(s) was done. The y-axis denotes the prevalence/probability, while the x-axis indicates the subject age. a) Early-onset mid-resolving eczema in studies [48, 74, 83, 90, 93] which presented trajectories by probability. b) Early-onset mid-resolving eczema in studies [65, 71] which presented trajectories by prevalence. c) Early-onset early-resolving eczema in studies [24, 61] which presented trajectories by probability.