RxAnte

The prediction is the original product and remains the thing plans buy: modeling which members will lapse before they lapse, from pharmacy and claims history, with prediction functions trained on historical claims as RxAnte's patents describe. That is a learned component, and the targeting does not exist without it.

The business has since acquired a second half that is not a model at all. The pharmacy service dispenses, packages, delivers and staffs pharmacist reviews, which is logistics and labor. The analytics depend on the model, and the pharmacy half of the business does not.

Low autonomy in the usual sense and appropriately so. The models rank members; pharmacists conduct the reviews, prescribers change the regimens, and members decide whether to enrol and whether to take anything.

The consequential automated act is selection rather than decision. Who appears on the outreach list determines who is offered intensive support, and that list is generated by a model. Nothing published describes whether a plan can override the ranking, whether clinicians can add members the score did not surface, or what happens to members who decline once and are then deprioritised.

The approach is published in RxAnte's patents, and what the model aims at is stated plainly. The Value of Future Adherence patent (US 10,318,897) describes a score built from three components, the probability a member becomes non adherent, the cost reduction associated with adherence and the probability of converting to adherence, with prediction functions trained on historical claims, compared across several statistical methods including ensemble methods and selected on held out validation data; the legal page lists the patents behind the core analytics, the VFA score and the RxEffect platform.

The product pages state the objective: identify the members with the greatest cost savings potential so a plan can engineer the return on its program, scored against named Star Ratings and HEDIS measure specifications. The features, the method finally selected, the versions in production and any accuracy figure are not published, so what is public shows how the score is constructed and what it aims at, not how well it hits.

The models are RxAnte's own, and its patents describe how they are built. The Value of Future Adherence patent (US 10,318,897) sets out a score combining the probability a member becomes non adherent, the cost reduction associated with adherence and the probability of converting to adherence, with prediction functions trained on historical claims, compared across several statistical methods including ensemble methods and selected on held out validation data.

No outside model provider sits between the member record and the score, so the chain is short. Everything around the model is unstated: no hosting arrangement, no subprocessor list, no retention schedule and no secondary use position, for a holding that places a plan's full pharmacy and claims history across millions of members beside the home addresses and household context the pharmacy arm collects. The pooling question matters here because a targeting model gains from breadth: nothing states whether one plan's member data informs models serving another.

A matched design with a stated hypothesis, and short of independent.

The published analysis compares members enrolled in the pharmacy service against matched members who filled prescriptions elsewhere, across total cost of care, unplanned care events and adherence at six months, with a hypothesis stated in advance that savings would concentrate in the high scoring subgroup. A matched control group and a pre stated subgroup hypothesis are real methodological choices.

Three things weaken the evidence. The analysis is self published rather than peer reviewed. It was conducted within a health plan belonging to the same organization as the company's investor, so the study site is not arm's length. And matched controls are not randomization: members who accept an intensive pharmacy service differ from those who decline in ways matching cannot fully capture.

General assurances exist for the pharmacy and none for the analytics. Mosaic's Notice of Privacy Practices commits it to maintaining the privacy and security of member health information, to breach notification, and to no research use without consent. Nothing comparable is published for the analytics platform: no retention schedule, no encryption detail, no de identification practice and no statement of whether member data is used to build or refit the targeting models.

The data is a plan's complete pharmacy and claims history across millions of members, which reveals conditions a person may never have discussed with anyone else, and the pharmacy arm adds home delivery addresses and household context. Nothing is published on safety engineering either: how a member wrongly ranked, or wrongly skipped, is detected and corrected.

HIPAA status is documented for one half of the business. Mosaic Pharmacy Service, RxAnte's pharmacy subsidiary, publishes a Notice of Privacy Practices in English and Spanish stating that it is required by law to maintain the privacy and security of protected health information, will notify patients of a breach, will not use health information for research without consent, and limits marketing to what HIPAA allows.

The analytics arm, which runs the models on a plan's pharmacy and claims data, states no status: RxAnte's own privacy policy (last modified February 2023) covers website visitors only, and no business associate statement or agreement terms are published. The two relationships differ, analytics vendor to a plan on one side and a dispensing pharmacy with its own relationship to the member on the other, and which instrument covers which data flow is not described.

Two recognized certifications are named in RxAnte's own material, and neither carries a current date. Every page of rxante.com carries HITRUST Certified and AICPA SOC 2 marks in the footer. The dated evidence behind them is older: a November 2016 release announcing HITRUST CSF certification and SOC 2 Type II audits completed for two consecutive years, and a 2018 release describing its systems as HITRUST and SOC 2 certified. No report, certificate date, audit period or scope is published. Operating the licensed Mosaic pharmacy alongside the analytics means two control environments, and nothing published says whether one attestation reaches both.

No device pathway applies and none is claimed.

The regulatory machinery that shapes this product is the Medicare Star Ratings system, because it defines what the models chase. Medication adherence measures carry triple weight and bear on roughly half of a plan's total star performance, and star performance determines bonus payments. So the commercial value of this product is set by a federal quality program rather than by a clinical outcome.

The pharmacy arm sits under an entirely different regime: state pharmacy licensure, dispensing rules and mail order requirements. The combined offering is a regulated dispensing operation as well as software.

No subgroup performance, monitoring policy or fairness analysis is published, and the objective function itself is the governance question.

The score ranks members by the cost their non adherence would generate. That is a legitimate way to run a program with finite pharmacists, and it is not the same as ranking by clinical need. A member whose lapse is dangerous but inexpensive ranks below one whose lapse is costly, and the two are not the same people. The company names the criterion openly rather than describing it as risk, so the trade it is making is visible.

The measure set narrows it further. The adherence measures that drive star performance cover a small number of drug classes, so a member non adherent to a medication outside those classes generates no measurable value and is correspondingly less likely to be surfaced.

The patents describe validating the models on held out historical data, a whole population test. Nothing published addresses whether ranking, outreach, enrollment or outcomes differ by race, income, language or rurality, in a program aimed explicitly at medically complex and vulnerable members.

People act on every output. The models rank members; pharmacists conduct the reviews, prescribers change regimens, and members decide whether to enroll, so no one's medication changes on a score alone. The objective is stated openly: the score identifies members with the greatest cost savings potential.

A ranking built on savings potential diverges from one built on clinical need in a predictable direction: a member with serious medication problems and little scope for cost reduction ranks below one whose adherence gap is expensive to the plan. Nothing stands behind the ranking itself. No accuracy or calibration figure is published for the score, no peer reviewed error rate is cited, and no warranty, indemnity or remediation commitment covers a member the model ranked low who needed help.

Electronic health record integration is not claimed for the analytics: the models run on plan side pharmacy and claims data, and the outputs go to plan quality teams, to the company's own pharmacists and to providers through its RxEffect workflow platform. Two connections are claimed without a system named.

Mosaic states that it completes comprehensive medication reviews in the plan's medication therapy management vendor platform, and that it coordinates regimen changes with prescribers and caregivers. How that prescriber contact happens, and whether anything reaches the prescriber's own record system rather than arriving as a call or a fax, is not described.

RxAnte runs its analytics as a hosted service: plans send pharmacy and claims data, and pharmacists, care managers and community providers work in its RxEffect platform. The customers are United States health plans in the Medicare program, so the location is implied rather than committed. No hosting provider, region, residency commitment or retention position is published.

The pharmacy operation has a physical footprint the analytics does not, since medications are dispensed, packaged and shipped from a licensed facility, which matters for a plan evaluating national coverage.

No price, mechanism or unit of sale is published for the analytics, and nothing shows what a plan pays for the pharmacy service. The one cost statement is on the member side: Mosaic describes its delivery and support as coming at no added cost to the member beyond the usual copay, with medications billed through the member's coverage. The structure itself raises a question.

The company's analytics identify which members would benefit most from intensive pharmacy support, and the company also owns the pharmacy that provides it, so a vendor is scoring demand for its own service. Nothing published says whether the analytics can be bought without the pharmacy, whether the score behaves the same way for plans that use a different pharmacy, or how each side is priced, and a per member analytics fee and a dispensing margin are different commercial objects.

Concentrated on one buyer and deep within it. The center of gravity is Medicare Advantage prescription drug plans, where the star measures create the strongest incentive, with coverage extending to HEDIS and patient experience measures for other lines.

Stated reach is substantial: more than 14 million lives under the analytics, and a claim to manage Part D star performance for 70 percent of the lives in five star plans nationally. Clinically the scope follows the star measures, so a small number of chronic medication classes dominate, while the pharmacy service targets medically complex members on multiple daily medications. United States only, which is inherent since the product exists because of how this quality and payment system works.