Methodology: How does this all work?

HIV-ASSIST utilizes a multi-criteria decision analysis framework to rank a large number of 2, 3, and 4 drug  combinations based on specific patient attributes (drug resistance, comorbidities, comedications, treatment history, viral load, CD4, tropism, HLA status).  HIV-ASSIST algorithms result in an aggregate  "Weighted Utility Score," constructed on a scale of 0-10+ (or higher in some case), where lower numbers are more 'preferred' regimens.  The outcome of interest is an aggregate of likelihood of viral suppression (e.g., drug activity) and tolerability (e.g., minimize drug interactions, impact on comorbidities, pill burden).  When interpreting results, clinicians should review the 'Rationale' tab for each proposed regimen and assess whether the individual mathematical 'penalties' and 'prioritizations' align with their assessment of an individual case.  For example, because the HIV-ASSIST score is 'weighing' both tolerability and activity, there may be instances in which a 'more tolerable' regimen (e.g., less pills) is ranked higher than one with a greater number of active drugs (e.g., more likely to be effective for goal of viral suppression).  All results and decision making steps are presented for clinicians to consider their own relative preferences for these individual factors that influence decision making.

Here's how we get the Weighted Score:

1. The 'Base-Score': the starting point
  • Our rankings utilize the current Department of Health and Human Services (DHHS) and International Antiviral Society (IAS) guidelines. When there are discrepancies between the guidelines, our ranking system was based on internal expert opinion of the Scientific Adivsory Panel, with a goal of quarterly updates.
  • The 'base-score' is our starting point for each potential ARV regimen.
  • Our system assigns a 'base-score' or rank on the basis of the DHHS and IAS guidelines for Naive patients.
  • Our numeric ranking system loosely follows the DHHS format of 'recommended Initial Regimen for Most People' (1), 'Recommended Initial Regimen in Certain Clinical Situations' (2), and other regimens with some evidence (rank 3-6).
  • There are many drug permutations that have not been formally recommended or studied, including some NRTI-sparing regimens. Our expert panel ranked such regimens between 3 and 6 in our 'base-score,' on the quality of available evidence
2. Mutations
  • We first incorporate viral mutations based on the Stanford Database to ascertain the number of active drugs in a regimen and apply a mathematical penalty for regimens containing less than 3 active drugs
  • When a drug is resistant (i.e. HIV Stanford mutation penalty >60), all regimens that include drugs to which the virus is resistant are excluded from our final recommendations list (*Note: this does not apply to 3TC and FTC, which are occasionally retained despite resistance)
  • ACTIVE DRUGS: our final report lists the number of fully active drugs in the regimen, and this factors into the final 'weighted score'
3. Comorbidities and Side-Effects
  • We recognize that comorbidities and side-effects are important considerations in deciding an ARV regimen
  • For each of the listed comorbidities, we evaluate each individual drug and assign a score of 0-2, in which 0 represents 'no issues' and 2 represents contraindications
  • We eliminate all ARV drug combinations in which one of the component drugs has a contraindication.
  • For all other ARV combinations, we incorporate a 'mathematical penalty' that is the SUM of all comorbidity scores for individual drugs in that regimen.
4. Drug Interactions
  • Disclaimer: We are not a comprehensive pharmacy database or interaction checker. We evaluate only one-to-one interactions, based on available  data in the University of Liverpool HIV Drug interaction checker, and in the drug interaction tables in the DHHS guidelines. We do not evaluate multi-way interactions.
  • For each listed comedication, we evaluate each individual drug and assign a score of 0-2, in which 0 represents 'no interactions' and 2 represents contraindications
  • We eliminate all ARV drug combinations in which one of the component drugs has a contraindication with a selected comedication.
  • For all other ARV combinations, we incorporate a 'mathematical penalty' that is the SUM of all drug interaction scores for individual drugs in that regimen.
5. Poor Adherence
  • Patients with poor adherence may need special considerations 
  • As such, we incorporate two sets of preferences for individuals with poor adherence
  • First, if selected, we preferentially prioritize (mathematically) regimens with at least 3 fully active drugs, and prioritize regimens considered to have a higher barrier to resistance
  • Second, if selected, we preferentially prioritize (mathematically) regimens with lesser pill burden and easier administration frequency. 
  • *Note, the degree to which clinicians may weigh these factors may differ for specific individuals. As such, we display the number of fully active agents and pill burden and dosing frequency on the main output page to allow additional clinicain considerations
6. Pill burden 
  • Our weighted score additionally incorporates preferences for fewer pills and simpler dosing frequency. *Note, the degree to which clinicians may weigh  pill burden and dosing frequency may differ; these considerations are displayed on the main output page to allow additional clinicain considerations
7. Other factors
  • There are over 30 FDA approved ARV agents in multiple mechanistic classes, with over 2000 potential 1 drug, 2 drug, 3 drug, and 4 drug combinations
  • We DO NOT evaluate all potential combinations
  • Regimens with a single agent are generally excluded in non-suppressed patients (e.g. PI monotherapy)
  • Regimens with multiple drugs from the same drug class are excluded (except for NRTIs) (e.g. dual PI regimens)
  • We do not consider T-20.  Emerging drugs and combinations are added semi-annually.
  • We display a list of ARV's to exclude from consideration, which can be modified by the clinician. In general, we automatically exclude older NRTIs and PIs.  We also exclude TDF regimens in favor of TAF, but this can be adjusted by users. 
  • In general, our algorithm evaluates but does not display RAL regimens when a similar regimen with DTG is available