China
Africa
Explore chapters and articles related to this topic
Healthcare as Complex, Entropic and Ethical
Published in Lesley Kuhn, Kieran Le Plastrier, Managing Complexity in Healthcare, 2022
Lesley Kuhn, Kieran Le Plastrier
Placing the relationship between patients and healthcare practitioners as central to healthcare has applied implications for how healthcare is managed and strategies for improvement. For example, this model suggests that interventions that have a detrimental effect on the patient–practitioner relationship are likely to have profound detrimental non-linear effects on patient health and on the practice of healthcare. Interventions that place pressure on practitioners will likely be detrimental, as a practitioner feeling pressured, is experiencing raised levels of entropy and their capacity to hold and absorb the entropy expelled by the suffering patient will be impeded. Conversely, interventions that contribute to minimising entropy across all of the stakeholders can be understood to nudge the practice of healthcare towards more optimal outcomes.
Statistical Methods for Big Data
Published in Ching-Yu Cheng, Tien Yin Wong, Ophthalmic Epidemiology, 2022
Emmanuel Kemel, Alexandre Thiery, Simon Nusinovici
To model the relationships between an outcome variable Y and predictor X, the logistic model uses a sigmoid transformation (or logit function) that can be written as log(p(Y)/(1 – p(Y)) = α + βX. A positive (negative) coefficient β corresponds to a positive (negative) association between the predictors and the outcome. The exponential of the coefficients can be interpreted as the odds ratios. The interpretability of odds ratios is the main advantage of the logistic model. Standard errors of the coefficients are also estimated, which allows confidence intervals to be calculated and a hypothesis to be tested. Additionally, non-linear effects and interactions can be added. Indeed, the linearity assumption can be relaxed using different methods (10), such as polynomial components, smoothing splines, or generalized additive models (GAM). The last option provides a general framework for extending linear modeling by allowing non-linear functions of each variable (11) (Figure 5.2A).
Can’t the computer just take care of all of this?
Published in Thomas A. Gerds, Michael W. Kattan, Medical Risk Prediction, 2021
Thomas A. Gerds, Michael W. Kattan
Another example of an unsupervised modeling step is to use restricted cubic splines for a continuous predictor variable (Section 4.3.2). This is a very useful technique for allowing a continuous variable to have a non-linear effect. For example, many routine blood tests have a normal range such that either low or high values may indicate high risk. The challenges here are which continuous variables to add the splines to and how many knots to use for each. More knots are needed to describe more flexible relationships. Since we are unsupervised, we must rely upon quantiles of the distribution for knot placement or use an expert's opinion regarding the biology in order to choose the number and/or placement of knots. However, if the expert indicates a very complicated relationship and hence suggests many knots, the sample size may still limit the feasibility. In practice, simply using 3 knots placed at the quartiles works pretty well.
Age-specific survival trends and life-years lost in women with breast cancer 1990–2016: the NORDCAN survival studies
Published in Acta Oncologica, 2022
Frida E. Lundberg, Niels Kroman, Mats Lambe, Therese M.-L. Andersson, Gerda Engholm, Tom Børge Johannesen, Anni Virtanen, David Pettersson, Elínborg J. Ólafsdóttir, Helgi Birgisson, Paul C. Lambert, Lina Steinrud Mørch, Anna L. V. Johansson
The flexible parametric relative survival models used restricted cubic splines with 5 degrees of freedom (df) for the log cumulative baseline excess hazard over time since diagnosis [9]. We included age and calendar year at diagnosis as continuous non-linear effects using restricted cubic splines with 3 df, and with 2 df for their two-way interaction. A three-way interaction between age, calendar year, and time-since-diagnosis was included as time-dependent splines with 3 df for each time-dependent effect (i.e., relaxing the proportional excess hazards assumption). For Iceland, we used simplified models excluding the three-way interaction and 2 df for the time-dependent effects. To improve model stability, 96% of the age distribution was modeled continuously while individuals outside the 2nd and 98th percentile of age had their age reassigned to those percentile limits and were assumed to have the same relative survival, i.e., winsorizing [10].
Latent triple trajectories of substance use as predictors for the onset of antisocial personality disorder among urban African American and Puerto Rican adults: A 22-year longitudinal study
Published in Substance Abuse, 2022
Jung Yeon Lee, Kerstin Pahl, Wonkuk Kim
The present study is unique because we examined the trajectories of alcohol, cigarette, cannabis use from mid adolescence through the emerging adulthood, risk taking and rule breaking behaviors in mid adolescence, and physical and sexual victimization in late adolescence to predict the onset of ASPD during the period between emerging adulthood and adulthood among African Americans and Latinxs. Also, it allows us to estimate the hazard of a specific event such as ASPD, since we used the interval censored Cox Proportional Hazard Model for investigating the effect of several variables upon a time specified event (i.e., ASPD) takes to happen. Practically, the structure (e.g., linear or non-linear) of substance use variables is not known to have effects on the outcome variable (e.g., ASPD); nonetheless, our approach can handle some non-linear effects without specifying the non-linear way of association.
Accurate population-based model for individual prediction of colon cancer recurrence
Published in Acta Oncologica, 2021
E. Osterman, J. Ekström, T. Sjöblom, H. Kørner, T. Å. Myklebust, M. G. Guren, B. Glimelius
Unadjusted HRs and confidence intervals are presented in Supplementary Table 3. The multivariable model was based on previous research, guidelines, known collinearity (e.g., for positive nodes and pN-stage), and the univariable analyses. Several ways to represent node status and risk factors were assessed in adjusted models. Perforation and malignancy grades did not contribute much to the model and were removed in favour of simplicity, HR 0.97 and 0.89 with non-significant p-values. Adjuvant therapy was kept in the model to allow for adjustment. The multivariable model was calculated from 7048 cases after exclusion of patients with missing data (mainly pT3 substage, vascular and perineural invasion). It performed well when tested in model data and is presented in Table 2. The model includes restricted cubic splines to model non-linear effects of both found and positive nodes and the increased risk of missed nodes with low node yields without introducing categorical variables. The non-linear effects are visualised in Supplementary Figure 1 in the form of effect plots. The calibration curves for the model, internal and external validation and other nomograms are presented in Figure 2. The AUC was 0.78 (95% CI 0.76–0.79) in the model data and 0.76 (95% CI 0.74–0.78) in the internal validation data.