Design and use of a gender-neutral, individualized wrinkle prediction model
383
Presented by: Ken Suematsu
Background and Objectives
The degree and rate of wrinkle progression is greatly affected by an individual's skin characteristics. In previous research, we developed a multivariate prediction model for determining the future state of wrinkles based on the longitudinal progression of wrinkles among individual subjects. The proposed model was able to predict wrinkles with high accuracy in middle-aged Japanese women, suggesting that it could serve as a useful guide for optimizing individualized wrinkle prevention strategies. However, validation of the results using an external study population was quite limited. The objectives of the current study are 1) to expand the generalizability of our wrinkle prediction model by including new external validation data from male subjects, and 2) to develop a practical customer-focused tool that utilizes our wrinkle predication model.
Methods
The new external validation data for the wrinkle prediction model were measurements of skin conditions and wrinkle assessment values obtained from healthy Japanese men in 2017 and 2021. To update the wrinkle prediction model, we adopted the intercept adjustment method and the intercept and slope adjustment methods, which are the most popular methods recommended in international guidelines for clinical prediction model research1. The apparent performance of the updated prediction model was assessed using R-squared and root mean squared error. Additionally, the prediction accuracy of wrinkles was evaluated using participant data collected after four years. Finally, as a practical customer-focused tool of the wrinkle prediction model, we developed a web application for easy online wrinkle prediction.
Results and discussion
The prediction score of the original wrinkle prediction model was re-calculated after inclusion of the data of Japanese males (n=85) prepared for external validation. Then, parameter estimation for the intercept and slope was conducted. Results showed that the apparent performance of the revised wrinkle prediction model had an explained variance ratio of 0.63 (95% confidence interval: 0.56-0.70) and root mean square error of 0.88 (95% confidence interval: 0.76-0.90). These values indicate that the model could predict the new population with good accuracy. To evaluate the test performance of the updated wrinkle prediction model, it was used to predict the wrinkles of subjects (n=25) after 4 years. When compared against the actual wrinkle measures taken from subjects, the results showed that the explained variance ratio was 0.60 (95% confidence interval: 0.54-0.66) and the root mean square error was 1.07 (95% confidence interval: 0.83-1.31). This result indicate that the model could be successfully used for gender-neutral wrinkle prediction. A final wrinkle prediction model was formulated by re-calibrating the intercept and overall slope, while maintaining the predictors of age, sebum, skin color a* value, skin color L* value, and the interaction term between sebum and skin color a* value. For the practical customer-focused tool of the wrinkle prediction model, a non-contact online service was created that utilized the model in combination with an AI skin analysis function provided by Perfect Corporation. In the new service, the skin color parameter has been replaced with the color information of the user's face image taken by the AI skin analysis function. Also, the sebum volume has been replaced with self-assessed standard values at four levels. In addition to providing a predicted profile of an individualized future wrinkle level, the new service displays factors that may lead to more severe wrinkles and their level of risk based on the observed predictors for each individual.
Conclusion
The wrinkle predication model presented here improves on our existing model that was based on data from middle-aged Japanese woman using age and only three skin surface conditions to a model that is not limited by gender. We have also incorporated the wrinkle prediction model into a practical user-focused tool and shared it with consumers. The wrinkle prediction tool developed in this study can be used to help consumers make better choices about wrinkle care and motivate them to voluntarily prevent wrinkles.
1. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): the TRIPOD statement. Ann Intern Med. 2015;162(1):55-63.
The degree and rate of wrinkle progression is greatly affected by an individual's skin characteristics. In previous research, we developed a multivariate prediction model for determining the future state of wrinkles based on the longitudinal progression of wrinkles among individual subjects. The proposed model was able to predict wrinkles with high accuracy in middle-aged Japanese women, suggesting that it could serve as a useful guide for optimizing individualized wrinkle prevention strategies. However, validation of the results using an external study population was quite limited. The objectives of the current study are 1) to expand the generalizability of our wrinkle prediction model by including new external validation data from male subjects, and 2) to develop a practical customer-focused tool that utilizes our wrinkle predication model.
Methods
The new external validation data for the wrinkle prediction model were measurements of skin conditions and wrinkle assessment values obtained from healthy Japanese men in 2017 and 2021. To update the wrinkle prediction model, we adopted the intercept adjustment method and the intercept and slope adjustment methods, which are the most popular methods recommended in international guidelines for clinical prediction model research1. The apparent performance of the updated prediction model was assessed using R-squared and root mean squared error. Additionally, the prediction accuracy of wrinkles was evaluated using participant data collected after four years. Finally, as a practical customer-focused tool of the wrinkle prediction model, we developed a web application for easy online wrinkle prediction.
Results and discussion
The prediction score of the original wrinkle prediction model was re-calculated after inclusion of the data of Japanese males (n=85) prepared for external validation. Then, parameter estimation for the intercept and slope was conducted. Results showed that the apparent performance of the revised wrinkle prediction model had an explained variance ratio of 0.63 (95% confidence interval: 0.56-0.70) and root mean square error of 0.88 (95% confidence interval: 0.76-0.90). These values indicate that the model could predict the new population with good accuracy. To evaluate the test performance of the updated wrinkle prediction model, it was used to predict the wrinkles of subjects (n=25) after 4 years. When compared against the actual wrinkle measures taken from subjects, the results showed that the explained variance ratio was 0.60 (95% confidence interval: 0.54-0.66) and the root mean square error was 1.07 (95% confidence interval: 0.83-1.31). This result indicate that the model could be successfully used for gender-neutral wrinkle prediction. A final wrinkle prediction model was formulated by re-calibrating the intercept and overall slope, while maintaining the predictors of age, sebum, skin color a* value, skin color L* value, and the interaction term between sebum and skin color a* value. For the practical customer-focused tool of the wrinkle prediction model, a non-contact online service was created that utilized the model in combination with an AI skin analysis function provided by Perfect Corporation. In the new service, the skin color parameter has been replaced with the color information of the user's face image taken by the AI skin analysis function. Also, the sebum volume has been replaced with self-assessed standard values at four levels. In addition to providing a predicted profile of an individualized future wrinkle level, the new service displays factors that may lead to more severe wrinkles and their level of risk based on the observed predictors for each individual.
Conclusion
The wrinkle predication model presented here improves on our existing model that was based on data from middle-aged Japanese woman using age and only three skin surface conditions to a model that is not limited by gender. We have also incorporated the wrinkle prediction model into a practical user-focused tool and shared it with consumers. The wrinkle prediction tool developed in this study can be used to help consumers make better choices about wrinkle care and motivate them to voluntarily prevent wrinkles.
1. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): the TRIPOD statement. Ann Intern Med. 2015;162(1):55-63.