Non-Invasive Screening for Sickle Cell Disease Using Only a Smartphone Application

Abstract

Introduction Pediatric mortality due to Sickle Cell Disease (SCD) remains high and often occurs prior to diagnosis in most black and brown low- and middle-income countries (LMICs), due to the high prevalence of SCD and limited resources in these countries.To that end, we developed a low-cost non-invasive screening tool that can be widely available at the point of need. Our screening tool utilizes the power of both image analysis and predictive algorithms as a novel approach to SCD screening and diagnostics. In this new app, we have combined the quantitative analysis of fingernail pallor, an adaptation of our previously described smartphone app for estimating blood hemoglobin levels using smartphone images of the nail beds, with qualitative symptom and medical history data to compute an SCD Likelihood Score (SLS) that provides an individual's pre-diagnostic test potential for disease, a key metric in the disease screening and diagnostic process. Methods 417 subjects ages 0.5-20 years old were enrolled from Children's Healthcare of Atlanta (CHOA) hematology/oncology clinic; of these patients, 219 patients had SCD, and 198 did not. Pediatric patients were screened because they best simulated the target populations aimed for the app's use. Our analysis was restricted to the presence of specific pain, past stroke history and/or symptoms, splenic crises, neurodevelopment disorder, family history of SCD, due to their specificity to SCD and their ability to be answered without the need for resource-intensive diagnostics. The SLS was calculated by determining the presence of these medical history items and the patient's anemia status informed by the patient's most recent CBC. For each symptom a weighted score was generated by finding the difference between the percentage of subjects with SCD and those without SCD, who experienced the symptom. The presence of symptoms was binarized with 1 indicating the presence of a symptom and 0 indicating the absence of a symptom. The binary value of the symptom was then multiplied by the symptom's respective weighted value. These values were totaled and normalized to give the overall clinical profile-weighted score with the same weight as the anemia status. A receiver operating characteristic analysis was conducted to select a threshold for the SLS that provided the most accurate diagnosis in our training dataset. We incorporated this SLS model into our lab's previously described smartphone app predictive algorithm . Results A clinical assessment of the screening app was conducted at CHOA and enrolled 68 participants from both in-patient and out-patient settings. Patient hemoglobin measurements were obtained using our previously described hemoglobin screening app and patients (and their parents for those needing their assistance) were asked to provide responses to the symptom survey. Of the five medical history items queried, family history and a history of splenic crisis when present was found to contribute to 30% and 41%, respectively, of the overall clinical profile score. Whereas the presence of a history of pain, stroke history, and learning difficulties made up smaller portions of this portion of the SLS. The app performed with a sensitivity of 100% and a specificity of 75% in screening for SCD in patients <5 years of age, the target demographic for early screening. Conclusion This SCD screening application has the potential to expand and improve SCD outcomes in LMICs where access to the gold standard or any accurate diagnostic tools is limited. The app's non-invasive and easy-to-use nature and the ubiquity of smartphones globally make this screening tool a practical and feasible mode of screening. Early disease screening is paramount to survival for children with SCD and if this tool proves to be accurate in these settings, can be lifesaving for the hundreds of thousands of infants with SCD born in LMICs per year. Non-invasive modalities such as our SCD screening application can enable resource allocation and cost reduction by identifying those patients who do and do not have a high likelihood of disease, thus reserving tests for those who are high risk.