Why BGPT?
logo

Review Claim by Claim

Check each statement against experiments, exact results, and limitations, with provenance intact.Know what the science actually supports before you trust the answer.

Press Enter ↵ to review paper



    Explore by Goal












     Quick Answer


    What this paper adds (in plain terms): Using UK primary-care records (CPRD Aurum; 1998–2018), it reports an exponential rise in new/incident recorded autism diagnoses, with the largest growth occurring among adults and among females over time.



     Long Answer


    Paper Review (Skeptical, evidence-based): Time trends in autism diagnosis over 20 years: a UK population-based cohort study

    DOI: 10.1111/jcpp.13505 • Design: Population-based observational cohort using CPRD primary care records • Primary question: Does the incident recorded autism diagnosis rate increase in the UK over 1998–2018?

    1) Visuals first: what the paper claims

    The curve above is a reconstruction from the paper’s reported exponential model fit (incidence index relative to 1998=100; exponentiated coefficient ~1.07; R²=0.98; 1998–2018).
    This is not the full-year series; it only uses male-share values explicitly visible in the provided Table 1 excerpt (1998, 2005, 2010, 2015, 2018).

    2) Methods: what was measured, and what can’t be concluded

    Data source & cohort. CPRD Aurum primary-care records from 738 GP practices in England (10% of practices) and Northern Ireland, including diagnoses coded over time.
    CPRD’s representativeness and quality checks are described via the CPRD data resource profile.
    Outcome definition: “incident recorded autism diagnosis,” not true incidence of autism symptoms. The paper defines the yearly “incidence” as the percentage of “acceptable” patients who receive a new record of an autism code in that year.
    • Key inference constraint: Because CPRD captures diagnoses that enter primary care coding (and may miss diagnoses made elsewhere), the paper’s main conclusion targets recorded diagnostic activity, not population changes in underlying autism traits.
    • Coding drift across DSM revisions: The BA/SA proxy uses older subcodes; after DSM-5 revisions (2013) some labels were integrated into ASD coding, constraining interpretation of SA vs BA shifts after that time.
    • Validation limitations: Autism code validity is supported by validation work (positive predictive value), but validation generally compares with clinical records rather than population-wide ascertainment of autism traits.

    3) Results: strongest signals and how to interpret them

    Overall trend (incident recorded autism diagnoses): The paper reports an overall 787% increase in recorded incidence from 1998 to 2018 with an exponential model fit (R²=0.98; exponentiated coefficient 1.07, 95% CI [1.06, 1.08]; p<.001).
    Moderation by age band (where diagnosis is occurring): The reported multivariable coefficients (coefficient reported as exponentiated relative to preschool=reference) show larger increases for childhood, adolescence, and especially adulthood, relative to preschool.
    Moderation by gender (females rising faster): Female gender significantly predicted a larger increase in incidence over time (exponentiated coefficient 1.02, 95% CI [1.01, 1.03], p<.001).
    This is consistent with broader evidence that diagnosed autism sex ratios can shift with changing recognition and diagnostic practices, including under-identification of girls/women in earlier periods.
    BA vs SA proxy: BA diagnoses increased more than SA until ~2013, after which DSM-5 coding changes restricted the BA/SA classification.

    4) Critical appraisal: what would falsify the paper’s main interpretation?

    What is strong:
    • Scale + representativeness (for primary care capture): Millions of patients in denominators over time, with representativeness considerations for CPRD practices described in data resource documentation.
    • Analytical transparency about encoding: The paper explicitly defines its incident diagnosis measure, discusses denominator construction (“acceptable” patients), and discusses DSM coding constraints affecting sub-analyses.
    Key limitations / blind spots that could generate an apparent increase in recorded incidence:
    • Private-sector under-ascertainment: If families increasingly used private diagnostic routes and those diagnoses fail to enter GP coding, trends could move in either direction depending on how recording practices change. The authors argue CPRD likely underreports autism diagnoses overall, which complicates translation of recorded incidence into symptom prevalence.
    • Diagnostic substitution & coding drift: Even with high PPV, “diagnosis labels” can broaden/shrink (DSM transitions, clinical awareness), producing secular changes in recorded coding without changes in underlying traits. The paper itself aligns with this possibility (including threshold shifts) but cannot adjudicate symptom-level causation from code-level incidence.
    • Confounding by service accessibility and policy changes: The authors mention NHS waiting times and policy directives (e.g., Adult autism services statutory duties). However, observational designs without direct measurement of referral/access and coding changes can’t fully isolate which mechanisms drive the trend.
    Replicability considerations: CPRD is restricted under license, which constrains independent re-analysis by the broader community; however, the statistical approach and code variables are described at a level that can support methodological replication (with access).

    5) Practical takeaways (careful wording: what is and isn’t shown)

    • Supported by this paper: In CPRD primary-care data, there is a large increase in newly recorded autism diagnoses between 1998 and 2018, and the increase is stronger for older age bands and for females.
    • Not uniquely determined: Whether underlying autism traits/symptoms truly increased in the population is not identifiable from the administrative trend alone; the study discusses plausible recognition/coding mechanisms but cannot measure them directly.
    • Interpretability hazard: Subcode-based “severity” (BA vs SA) is partially confounded by DSM-5 coding transitions, and only a fraction of cases could be classified for BA/SA, limiting certainty about severity composition changes.


    Feedback:    

    Updated: March 24, 2026

     BGPT Paper Review


    Study Novelty

    70%

    Moderately novel because it extends prior UK CPRD work by quantifying incident trends over a longer window (1998–2018) and stratifying by developmental stage, gender, and a severity proxy; however, it remains an administrative-data epidemiology study rather than a new causal mechanism.


    Scientific Quality

    80%

    Scientific quality is relatively high for an observational administrative study: large denominators, explicit outcome definition, formal modelling of time trends, and clear discussion of under-ascertainment and DSM coding constraints. Main quality risk: data access limits replication and the outcome is recorded diagnosis rather than direct symptom measurement.


    Study Generality

    60%

    Results generalize reasonably to UK primary-care coding contexts, but may be less generalizable to systems with different referral pathways, private-sector feeding-back of diagnoses, or different coding practices; severity proxy limitations further constrain cross-context comparisons.


    Study Usefulness

    80%

    Useful for policy and services planning because it quantifies where recorded diagnoses are rising (age and gender patterns). However, it cannot answer whether autism symptom prevalence truly increased without integrating symptom-level datasets.


    Study Reproducibility

    60%

    Reproducibility is methodologically moderate but constrained by restricted CPRD data licensing and lack of openly deposited analytic code in the provided text.


    Explanatory Depth

    70%

    Explanations focus on recognition/recording dynamics (adult service policy impacts, gendered identification issues, and diagnostic coding changes). While plausible, the paper does not directly measure recognition processes, service capacity, or changes in diagnostic thresholds, so mechanistic attribution remains partially inferential.


    🎁 Authors: Collect 225 Free Science Tokens (≈ $22.5 USD)

    Claim My Author Tokens

    Use for 56 days of free BGPT access (4 tokens = 1 day) or trade/sell (≈ $22.5 USD)

     Top Data Sources ExportMCP


     Analysis Wizard



    Reconstructs the paper’s exponential incidence index curve (1998=100) using the reported year-wise exponentiated coefficient, then plots it for visual comparison with reported 787% increase.



     Hypothesis Graveyard


    A purely etiologic acceleration (e.g., increased biological incidence) as the sole driver becomes less plausible if severity-proxy and coding-label structure changes track DSM transitions and if under-ascertainment mechanisms differentially affect age/gender groups; this paper explicitly suggests recognition/coding influences.


    A single simple “better awareness increases referrals equally across all groups” explanation is weakened because the paper reports differential acceleration by developmental stage and by gender (female increases larger; adults increase stronger), implying more than uniform outreach.

     Science Art


    Paper Review: Time trends in autism diagnosis over 20 years: a UK population‐based cohort study Science Art

     Science Movie


    Make a narrated HD Science movie for this answer ($32 per minute)




     Discussion


    Stay current without chasing every paper.

    Know what changed, what holds up, and what remains uncertain. Every Friday. No ads.


    My BGPT






     Trending