Also available in: Español
The tool would have made it possible to establish a subgroup at high risk of acquiring HIV, constituting about 1% of the population
By Francesc Martínez
During a presentation at the IDWeek 2016 conference recently held in New Orleans (USA), US researchers have shown that a computerized medical records analysis algorithm could adequately identify those people who could benefit most from the use of pre-exposure prophylaxis (PrEP).
PrEP was approved in the USA in 2012 (see La Noticia del Día 19/07/2012 in Spanish language) and recently in Europe (see La Noticia del Día 12/09/2016 in Spanish language). Although several clinical trials have showed good preventive effectiveness rates, due to its cost to the health system and the possible side effects related to its usage, it is essential to make a very reasoned selection of the people in whom the balance between risks, benefits and cost is acceptable.
With the objective of establishing a computerized algorithm that allows the precise and simple detection of those who could most benefit from PrEP, the authors of the present study carried out a three step process:
- The extraction of all clinically relevant data from the electronic records of Atrius Health, a mutual private healthcare from the Boston (USA) area with more than 800,000 patients. These data included demographic variables, recent diagnoses, medication prescriptions, medical and / or surgical interventions and analytical results.
- Compared the characteristics of the 138 participants who became infected by HIV between 2006 and 2015 to those of 100 participants of similar profiles, but remained HIV-negative, who acted as control subjects. This comparison was made using statistical techniques of logistic regression modelling and machine learning with the aim of recognizing data patterns that could help predict new infections.
- Based on all the analyzed data, determine if, from the general umbrella group of participants, it was possible to identify a subgroup with candidates for the use of PrEP.
When comparing logistic regression statistical techniques – more conventional – with those of machine learning, the latter obtained better results predicting the new infections; and among the machine learning techniques used, Ridge regression was the one that obtained better results.
Observing the evaluated variables in detail, 6.5% of the people who became infected had undergone anal cytology tests (while this percentage was 0.1% in the control group). A 3.6% of people who acquired HIV had been prescribed penicillin G (an antibiotic frequently used to treat syphilis), while the percentage in the control group of people who received this prescription was below 0.1%. Also, 5.8% of people who acquired HIV had a history of diagnosis of gonorrhoea, something that was observed in only 0.1% of people in the control group.
The vast majority of participants in the study were classified by the computer algorithm as having low or very low risk of acquiring HIV. Finally, out of the over 800,000 participants included, a total 8,414 (which would represent 1.1% of the total) were considered candidates for PrEP.
Given the high cost of PrEP, considering that 1% of the total population is a candidate to receive it is perhaps a figure still too high to consider this as a cost-effective approach, but it could be a good starting point to continue fine tuning the model. In fact, given that it is an algorithm based on machine learning, as more data from a larger number of records is incorporated, the tool may be able to establish subgroups of PrEP candidates representing a smaller percentage of the population than the present model.
According to the researchers, the next step will be to optimize and validate the present algorithm in specialized sexual health centres with the aim of verifying that it is able to increase the preventive role of PrEP in these environments in a cost-effective way.
Source: Aidsmap / Prepared by the author (gTt).
References: Krakower D et al. Automated identification of potential candidates for HIV pre-exposure prophylaxis using electronic health record data. IDWeek, abstract 860, 2016.