Explore chapters and articles related to this topic
Machine learning and public health
Published in Sridhar Venkatapuram, Alex Broadbent, The Routledge Handbook of Philosophy of Public Health, 2023
Among the three types of data-science activities, causal inference enjoys a special status. First, causal inference is intimately linked to the self-understanding of epidemiology, as it is traditionally defined in terms of distribution and determinants of disease—with “determinants” taken to be causes (Broadbent 2013). Second, it is often assumed that causal inference is epistemically superior for important public health purposes, specifically when seeking to understand the outcome of a contemplated intervention. In establishing a causal link between cancer screening and a mortality reduction, policymakers and healthcare professionals are enabled not merely to foretell how things will go in the future, but to develop reliable preventive measures to change the future for the better. An accurate prediction without a causal inference could be the result of more than one causal relationship, including some that do not result in a change in the outcome of interest, as when one pushes the needle of a barometer in hopes of averting a storm.
Real-World Data and Real-World Evidence
Published in Wei Zhang, Fangrong Yan, Feng Chen, Shein-Chung Chow, Advanced Statistics in Regulatory Critical Clinical Initiatives, 2022
Counterfactual causal inference was one of the most important statistical ideas of the past 50 years, which has been commonly used in multiple areas such as econometrics and epidemiology. Based on the theoretical framework of counterfactual causal inference, many methods are used for causal inference in observational studies. In this section, we discussed several of these casual inference methods: propensity score, disease risk score, and instrumental variables.
The Form of Structural Equation Models
Published in Douglas D. Gunzler, Adam T. Perzynski, Adam C. Carle, Structural Equation Modeling for Health and Medicine, 2021
Douglas D. Gunzler, Adam T. Perzynski, Adam C. Carle
The pseudo-isolation condition is particularly important for causal inference [12]. Causal inference is a process for determining whether a cause influences an effect. Most of the causes of the given effects should be in the model in order to justify the condition of pseudo-isolation. Otherwise, most of the causes would likely be in the error term and lead to a correlation of the error with the explanatory variables in the model [1]. In a misspecified model that omitted important explanatory variables, the causal impact of the predictors included in the model (e.g. in the structural model) may be incorrectly estimated.
Short-term risk of periprocedural stroke relative to radial vs. femoral access: systematic review, meta-analysis, study sequential analysis and meta-regression of 2,188,047 real-world cardiac catheterizations
Published in Expert Review of Cardiovascular Therapy, 2023
Jan Tužil, Jan Matějka, Mamas A. Mamas, Tomáš Doležal
The assumption of unmeasured confounding is a fundamental concern of causal inference based on observational data. A recommended reporting standard for meta-analyses is to conduct a post-estimation sensitivity analysis to assess how strong a relationship would have to be between an unmeasured confounder and the treatment assignment, as well as between the unmeasured confounder and the outcome, to explain away an observed treatment effect [33,34]. In addition to the abovementioned sub-group analyses, we calculated e-values [35]. E-value characterizes the extent of bias which would be required, hypothetically, to shift the pooled estimate to the null [34], in our case the risk ratio expressing the association that an unmeasured confounder(s) would need to have with both the treatment assignment and the outcome to ‘explain away’ the observed treatment-outcome effect. E-values for the point estimate and the confidence range were calculated using Stata immediate command evalue assuming the relative outcomes on the risk ratio scale.
Parents Talking to Middle School Children about Sex: A Protective Factor against Suicide in Sexually Active Teens
Published in American Journal of Sexuality Education, 2023
Monica St. George, Danielle R. M. Niemela, Robert J. Zeglin
The results of this study should only be considered in light of the study’s limitations. First, the sample was drawn from only one county in the urban coastal southeast United States. This may limit its generalizability to other, particularly more rural, populations. Also, the study instrument only offered “male” and “female” as gender identity options and did not ask about sexual orientation at all. This severely limits the study’s ability to assess whether the identified significant associations differ for sexual and gender minority adolescents, a growing cohort of young adults (Laughlin, 2016; White et al., 2018). A third limitation, one inherent in a non-experimental cross-sectional study, is the lack of causal inference that can be made based on the study’s results. Finally, the study instrument did not ask about the quality or reception of the parent’s conversation about sex. Young adults with varying levels of sexual knowledge, comfort, and values or those with stronger or weaker relationships with their parents may have experienced the conversation about sex markedly differently. Similarly, knowing whether the child or the parent initiated the conversation about sex could prove to be a significant moderator in the relationship between conversations about sex and suicidality. Other examples of possible mediating or moderating variables are perceived parental limit-setting and timing of conversation relative to sexual debut. The present study was not able to assess for such differences in these and similar variables. Despite these limitations, the present study was able to answer its guiding research question.
Impact of COVID-19 pandemic on road safety in Tamil Nadu, India
Published in International Journal of Injury Control and Safety Promotion, 2022
Kandaswamy Paramasivan, Nandan Sudarsanam
This study has a few limitations, one of which is that the actual incidence of RTCs(ground-truth) is not reflected in the analysis due to the underreporting of RTCs, All four methods of counterfactual prediction used in this study revealed more or less consistent results, which boosted the confidence of the estimate of causal inference. However, if the data contains anomalous values or has rapidly fluctuating trends, then prediction accuracy will be negatively affected. Missing values in time-series data adversely affect the performance of all four popular models. It is also significant to note that these models work well only for short-term predictions, and are not suitable for long-term forecasts. They are also unsuitable for time-series data that exhibit complex patterns (Pra, 2020). The study’s predictions are based on the assumption that the white noise in the time series is Gaussian; this simplification, if not representative, could compromise the accuracy of the prediction. Lastly, these models cannot be effectively utilized by researchers lacking domain knowledge of the subject in question. Prediction accuracy depends to a great extent on the inputs provided by the researchers, e.g. including details such as the holiday effect on traffic data; therefore, subject expertise is a pre-requisite (Taylor & Letham, 2018).