Risk heatmaps as visual displays: Opening movie studios after the COVID-19 shutdown

Communicating health information with visual displays

Well-designed visual displays have the power to convey health messages in clear, effective ways to non-experts, including journalists, patients and policymakers. Poorly designed visual displays, however, can confuse and alienate recipients, undermining health messages. In this Perspective, we propose a structured framework for effective visual communication of health information, using case examples of three common communication tasks: comparing treatment options, interpreting test results, and evaluating risk scenarios. We also show simple, practical ways to evaluate a design's success and guide improvements. The proposed framework is grounded in research on health risk communication, visualization and decision science, as well as our experience in communicating health data.

Revisiting prediction of collapse in hip osteonecrosis with artificial intelligence and machine learning: a new approach for quantifying and ranking the contribution and association of factors for collapse

PURPOSE

This study proposes machine learning to analyze the risk factors of the collapse in patients with non-traumatic hip osteonecrosis of the femoral head.

METHODS

We collected data of 900 consecutive patients (634 males) with bilateral (428) or unilateral non-traumatic osteonecrosis diagnosed before collapse (at stage I or stage II). The follow-up was average five years (3 to 8 years). A total of 50 variables related to the osteonecrosis were included in the study. The osteonecroses were randomly divided into a training set (80%) and a validation set (20%) with a similar percentage of hips with collapse in the two groups. Machine learning (ML) algorithms were trained with the selected variables. Performance was evaluated and the different factors (variables) for collapse were ranked with Shapley values. The primary outcome was prediction of occurrence of collapse from automated inventory systems.

RESULTS

In this series of patients, the accuracy with machine learning for predicting collapse within three years follow-up was 81.2%. Accuracies for predicting collapse within six to 12-24 months were 54.2%, 67.3%, and 71.2%, respectively, demonstrating that the accuracy is lower for a prevision in the short term than for the mid-term. Despite none of the risk-factors alone achieving statistical significance for prediction, the system allowed ranking the different variables for risk of collapse. The highest risk factors for collapse were sickle cell disease, liver, and cardiac transplantation treated with corticosteroids, osteonecrosis volume > 50% of the femoral head. Cancer (such as leukemia), alcohol abuse, lupus erythematosus, Crohn's disease, pemphigus vulgaris treated with corticosteroids, and osteonecrosis volume between 40 and 50% were medium risk factors for collapse. Familial cluster of collapse, HIV infection, chronic renal failure, nephrotic syndrome, and renal transplantation, when treated with corticosteroids, stage II, osteonecrosis volume between 30 and 40%, chemotherapy, hip pain with VAS > 6, and collapse progression on the contralateral side, were also significant but lowest risk factors. A heat map is proposed to illustrate the ranking of the combinations of the different variables. The highest risk of collapse is obtained with association of various risks factors.

CONCLUSION

This study, for the first time, demonstrated prediction of collapse and ranking of factors for collapse with a machine learning system. This study also shows that collapse is due to a multifactorial risk factors.