IDENTITY
Our decentralized approach to patient tracking is in three parts. First, we focus on the relative matching of patients over hours, months or years by matching patient encounters to themselves, without the specific need to link to a national registry. For example, image 1 shows face image acquisition performed at an extended distance using a smart-acquisition capability, and image 2 shows automatic biometric matching of the face between the first and second dose of a simulated vaccination campaign.
Image 1
Second, we focus on the decentralized issuance of unique identifiers to enable rural or offline clinics to issue identifiers that they or other clinics or health workers elsewhere can use without being dependent on centralized coordination. Thirdly, our Federated Biometrics approach allows a clinic or even the patient themselves to be the custodian of their biometric which other clinics or health workers can then use for identification with the patient’s permission.
We have developed Patient Matching performance metrics that quantify these match results. The graph below shows on the vertical axis Patient Matching Performance as a percentage from 0 to 100, and on the horizontal axis is shown the number of patients within a population. Considering just name and date of birth matching, as the number of patients increases, then the number of name collisions, misspellings and other inefficiencies also increases, resulting in a reduction in Patient Matching performance as shown by the lower descending line. Patient Matching Performance needs to be near to 100% - marked in purple - so a clinician or health worker has sufficient confidence to make an intervention.
Image 2
In our approach, one or both of a decentrally-issued identifier and one or more biometrics can be used to perform patient matching.
Below is a result of a prototype system that was developed and tested on HIV and TB patients working with the NHLS in South Africa. In this case, only multimodal biometric matching (iris + fingerprints) was used for patient matching. The figure shows on the Y-axis a cumulative histogram from 0-100% of match scores for 935 authentic biometric matches shown in blue, and 305,217 impostor biometric tests shown in brown. On the X-axis is the distribution of match scores; the lower the score the better the match. The lack of overlap in the areas under each of the blue and brown lines shows that there were no deduplication errors.
Results from surveys performed on HIV & TB patients showed broad acceptance of the use of biometrics for improving the efficiency of retrieving their medical information.
