Limits...
Visualizing Risk Prediction Models.

Van Belle V, Van Calster B - PLoS ONE (2015)

Bottom Line: To boost the uptake of these models in clinical practice, it is important that end-users understand how the model works and can efficiently communicate its results.The proposed methods summarize risk prediction models and risk predictions for specific patients in an alternative way.These representations may facilitate communication between clinicians and patients.

View Article: PubMed Central - PubMed

Affiliation: KU Leuven, Department of Electrical Engineering (ESAT), STADIUS Center for Dynamical Systems, Signal Processing and Data Analytics, Leuven, Belgium; iMinds Medical IT, KU Leuven, Leuven, Belgium.

ABSTRACT

Objective: Risk prediction models can assist clinicians in making decisions. To boost the uptake of these models in clinical practice, it is important that end-users understand how the model works and can efficiently communicate its results. We introduce novel methods for interpretable model visualization.

Methods: The proposed visualization techniques are applied to two prediction models from the Framingham Heart Study for the prediction of intermittent claudication and stroke after atrial fibrillation. We represent models using color bars, and visualize the risk estimation process for a specific patient using patient-specific contribution charts.

Results: The color-based model representations provide users with an attractive tool to instantly gauge the relative importance of the predictors. The patient-specific representations allow users to understand the relative contribution of each predictor to the patient's estimated risk, potentially providing insightful information on which to base further patient management. Extensions towards non-linear models and interactions are illustrated on an artificial dataset.

Conclusion: The proposed methods summarize risk prediction models and risk predictions for specific patients in an alternative way. These representations may facilitate communication between clinicians and patients.

No MeSH data available.


Related in: MedlinePlus

Graphical representation of an artificial logistic model that includes non-linear functional forms and interaction effects.For each predictor or interaction, the range is indicated below or next to the color bars and color plots. The color indicates the contribution to the linear predictor corresponding to the predictor values. The colors are converted to points by means of the color legend at the right of the graph. The sum of all points, i.e. the score, is then converted to the estimated risk by means of the color bar at the bottom. The triangles/diamonds indicate the predictor values and the corresponding risk estimate for a specific patient. The dashed gray lines are used to show percentiles.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4503430&req=5

pone.0132614.g008: Graphical representation of an artificial logistic model that includes non-linear functional forms and interaction effects.For each predictor or interaction, the range is indicated below or next to the color bars and color plots. The color indicates the contribution to the linear predictor corresponding to the predictor values. The colors are converted to points by means of the color legend at the right of the graph. The sum of all points, i.e. the score, is then converted to the estimated risk by means of the color bar at the bottom. The triangles/diamonds indicate the predictor values and the corresponding risk estimate for a specific patient. The dashed gray lines are used to show percentiles.

Mentions: For non-linear models the predictor contributions become βpfp(xp), with fp(xp) a non-linear transformation of the predictor xp, such as logarithmic or power transformations or restricted cubic splines. A non-linear effect can be represented in the same way as before, where the color of each color bar will now represent βpfp(xp) instead of βpxp. The presented visualizations can be further extended to include interaction effects. For each interaction effect, an additional color plot is added to the model representation. For the patient-specific representations, a bar is added with lengthβp,qfp,q(xp,xq)−mini∈D(βp,qfp,q(xip,xiq)),(4)with βp,q the coefficient of the interaction and fp,q(xp, xq) a transformation on both predictors involved in the interaction. This approach is illustrated in Fig 8 for an artificial example with 4 predictors: gender (binary), age (continuous), smoker (binary) and a biomarker (continuous). The visualized model is a logistic regression model with z = -1 + β1 *gender + β2 *(age– 60)2 + β3 *smoker + β4 *biomarker + β5 *gender*((age-60)/10)3 + β6 *gender*smoker + β7 *(age-60)2*biomarker, with β1 = -2, β2 = 0.005, β3 = 3, β4 = -0.02, β5 = 0.3, β6 = -2.5, and β7 = 0.00005. In addition to linear main effects, this model contains a non-linear main effect for age, an interaction effect between gender and age, an interaction effect between gender and smoker and an interaction effect between age and biomarker. In addition to the model visualization, the 5th and 95th percentile (for continuous predictors) have been added in dashed gray lines in Fig 8. The predictor values for one specific patient (a 62 year old smoking woman with a biomarker level of 82 U/mL) together with the estimated risk are indicated by means of the triangles and diamonds. The contribution chart for this patient is given in S5 Fig, where a contribution for all relevant interaction effects is added in addition to the contributions of the main effects.


Visualizing Risk Prediction Models.

Van Belle V, Van Calster B - PLoS ONE (2015)

Graphical representation of an artificial logistic model that includes non-linear functional forms and interaction effects.For each predictor or interaction, the range is indicated below or next to the color bars and color plots. The color indicates the contribution to the linear predictor corresponding to the predictor values. The colors are converted to points by means of the color legend at the right of the graph. The sum of all points, i.e. the score, is then converted to the estimated risk by means of the color bar at the bottom. The triangles/diamonds indicate the predictor values and the corresponding risk estimate for a specific patient. The dashed gray lines are used to show percentiles.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4503430&req=5

pone.0132614.g008: Graphical representation of an artificial logistic model that includes non-linear functional forms and interaction effects.For each predictor or interaction, the range is indicated below or next to the color bars and color plots. The color indicates the contribution to the linear predictor corresponding to the predictor values. The colors are converted to points by means of the color legend at the right of the graph. The sum of all points, i.e. the score, is then converted to the estimated risk by means of the color bar at the bottom. The triangles/diamonds indicate the predictor values and the corresponding risk estimate for a specific patient. The dashed gray lines are used to show percentiles.
Mentions: For non-linear models the predictor contributions become βpfp(xp), with fp(xp) a non-linear transformation of the predictor xp, such as logarithmic or power transformations or restricted cubic splines. A non-linear effect can be represented in the same way as before, where the color of each color bar will now represent βpfp(xp) instead of βpxp. The presented visualizations can be further extended to include interaction effects. For each interaction effect, an additional color plot is added to the model representation. For the patient-specific representations, a bar is added with lengthβp,qfp,q(xp,xq)−mini∈D(βp,qfp,q(xip,xiq)),(4)with βp,q the coefficient of the interaction and fp,q(xp, xq) a transformation on both predictors involved in the interaction. This approach is illustrated in Fig 8 for an artificial example with 4 predictors: gender (binary), age (continuous), smoker (binary) and a biomarker (continuous). The visualized model is a logistic regression model with z = -1 + β1 *gender + β2 *(age– 60)2 + β3 *smoker + β4 *biomarker + β5 *gender*((age-60)/10)3 + β6 *gender*smoker + β7 *(age-60)2*biomarker, with β1 = -2, β2 = 0.005, β3 = 3, β4 = -0.02, β5 = 0.3, β6 = -2.5, and β7 = 0.00005. In addition to linear main effects, this model contains a non-linear main effect for age, an interaction effect between gender and age, an interaction effect between gender and smoker and an interaction effect between age and biomarker. In addition to the model visualization, the 5th and 95th percentile (for continuous predictors) have been added in dashed gray lines in Fig 8. The predictor values for one specific patient (a 62 year old smoking woman with a biomarker level of 82 U/mL) together with the estimated risk are indicated by means of the triangles and diamonds. The contribution chart for this patient is given in S5 Fig, where a contribution for all relevant interaction effects is added in addition to the contributions of the main effects.

Bottom Line: To boost the uptake of these models in clinical practice, it is important that end-users understand how the model works and can efficiently communicate its results.The proposed methods summarize risk prediction models and risk predictions for specific patients in an alternative way.These representations may facilitate communication between clinicians and patients.

View Article: PubMed Central - PubMed

Affiliation: KU Leuven, Department of Electrical Engineering (ESAT), STADIUS Center for Dynamical Systems, Signal Processing and Data Analytics, Leuven, Belgium; iMinds Medical IT, KU Leuven, Leuven, Belgium.

ABSTRACT

Objective: Risk prediction models can assist clinicians in making decisions. To boost the uptake of these models in clinical practice, it is important that end-users understand how the model works and can efficiently communicate its results. We introduce novel methods for interpretable model visualization.

Methods: The proposed visualization techniques are applied to two prediction models from the Framingham Heart Study for the prediction of intermittent claudication and stroke after atrial fibrillation. We represent models using color bars, and visualize the risk estimation process for a specific patient using patient-specific contribution charts.

Results: The color-based model representations provide users with an attractive tool to instantly gauge the relative importance of the predictors. The patient-specific representations allow users to understand the relative contribution of each predictor to the patient's estimated risk, potentially providing insightful information on which to base further patient management. Extensions towards non-linear models and interactions are illustrated on an artificial dataset.

Conclusion: The proposed methods summarize risk prediction models and risk predictions for specific patients in an alternative way. These representations may facilitate communication between clinicians and patients.

No MeSH data available.


Related in: MedlinePlus