Independent Researcher, Los Altos, California, United States of America.
PLoS One. 2021 Dec 2;16(12):e0260738. doi: 10.1371/journal.pone.0260738. eCollection 2021.
An unbiased, widely accepted estimate of the rate of occurrence of new cases of autism over time would facilitate progress in understanding the causes of autism. The same may also apply to other disorders. While incidence is a widely used measure of occurrence, birth prevalence-the proportion of each birth year cohort with the disorder-is the appropriate measure for disorders and diseases of early childhood. Studies of autism epidemiology commonly speculate that estimates showing strong increases in rate of autism cases result from an increase in diagnosis rates rather than a true increase in cases. Unfortunately, current methods are not sufficient to provide a definitive resolution to this controversy. Prominent experts have written that it is virtually impossible to solve. This paper presents a novel method, time-to-event birth prevalence estimation (TTEPE), to provide accurate estimates of birth prevalence properly adjusted for changing diagnostic factors. It addresses the shortcomings of prior methods. TTEPE is based on well-known time-to-event (survival) analysis techniques. A discrete survival process models the rates of incident diagnoses by birth year and age. Diagnostic factors drive the probability of diagnosis as a function of the year of diagnosis. TTEPE models changes in diagnostic criteria, which can modify the effective birth prevalence when new criteria take effect. TTEPE incorporates the development of diagnosable symptoms with age. General-purpose optimization software estimates all parameters, forming a non-linear regression. The paper specifies all assumptions underlying the analysis and explores potential deviations from assumptions and optional additional analyses. A simulation study shows that TTEPE produces accurate parameter estimates, including trends in both birth prevalence and the probability of diagnosis in the presence of sampling effects from finite populations. TTEPE provides high power to resolve small differences in parameter values by utilizing all available data points.
一个公正的、被广泛接受的新自闭症病例发生率的估计值，将有助于我们深入了解自闭症的病因。这同样适用于其他疾病。虽然发病率是一种常用的发病衡量指标，但发病流行率（每一年龄组出生队列中患有该疾病的比例）是衡量儿童早期疾病和障碍的恰当指标。自闭症流行病学研究通常推测，发病率的显著上升表明诊断率上升，而非病例真正增加。遗憾的是，目前的方法无法明确解决这一争议。一些权威专家表示，这几乎是不可能解决的。本文提出了一种新的方法，即“事件时间发病流行率估计（TTEPE）”，该方法可以通过调整不断变化的诊断因素，为我们提供准确的发病流行率估计值。它解决了先前方法的缺点。TTEPE 基于广为人知的事件时间（生存）分析技术。离散生存过程通过出生年份和年龄来模拟新诊断病例的发生率。诊断因素通过诊断年份来驱动诊断概率。TTEPE 可用于模拟诊断标准的变化，当新标准生效时，这些变化可以修改有效发病流行率。TTEPE 将可诊断症状的发展纳入其中。通用优化软件可以估计所有参数，形成非线性回归。本文详细说明了分析所依据的所有假设，并探讨了潜在的偏离假设和可选的附加分析。一项模拟研究表明，TTEPE 可以在存在有限人群抽样效应的情况下，对包括发病流行率和诊断概率趋势在内的所有参数进行准确估计。TTEPE 可以充分利用所有可用的数据点，对参数值的微小差异进行高灵敏度的检测。
