MassMutual's Newly Granted Patent Turns Medical Claims Into a Mortality Score

US Patent 12,682,400 was granted to Massachusetts Mutual Life Insurance Company on July 14, 2026 (USPTO, July 2026). It encodes an applicant's medical claim history as 70 GloVe embedding coordinates and feeds them into a random forest that outputs a relative mortality risk score.

It was trained on roughly 40,000 historical underwriting applications carrying about 21,000 distinct diagnostic codes. The claim language, not the grant date, is what carriers renting a mortality-scoring model should read closely.

Key Takeaways

US 12,682,400, granted July 14, 2026, reduces a claims history to 70 embedding coordinates and feeds those plus age into a random forest returning a relative mortality risk score.
Roughly 40,000 historical applications carrying about 21,000 diagnostic and 8,000 procedural codes form the training set, and the label appears to be the underwriter's assigned risk class rather than matured mortality experience.
Five patents since 2020 cover overlapping fluidless-mortality architecture, which raises freedom-to-operate questions for any vendor selling a competing claims-to-embedding-to-random-forest pipeline.
15% average mortality slippage against priced assumptions across accelerated underwriting programs, with individual programs ranging from 5% to more than 30%.
Claim 1 requires an explanation tied to the codes behind the minimum and maximum values. That answers the black-box objection and does not satisfy the proxy testing New York's Circular Letter No. 7 requires.

Patent Details

US 12,682,400 B1 did not arrive alone. It is the newest member of a predictive-modeling family MassMutual has been building since at least 2020, several members of which reuse overlapping language around a fluidless mortality module paired with a smoking-propensity model and a prescription-fills model.

Patent	Granted	What it covers
US 11,694,775 B1	July 4, 2023	Underwriting based on predictive modeling with excluded mortality risk factors
US 11,710,564 B1	July 25, 2023	Fluidless mortality, smoking-propensity, and prescription-fills modules trained on historical clinical data
US 11,983,777 B1	May 14, 2024	Underwriting estimator returning an immediate risk-class estimate with feature-attribution explanations
US 12,205,690 B1	February 25, 2025	Excluded risk factor predictive modeling, filed March 8, 2022
US 12,682,400 B1	July 14, 2026	Medical Claims Risk Score: GloVe embeddings plus random forest mortality scoring, filed August 26, 2022

What Claim 1 Actually Covers

The independent claims describe a specific pipeline rather than a general appeal to machine learning. For each historical applicant record, the system feeds ICD diagnosis codes and CPT procedure codes from a third-party claims source into a word embedding model trained with GloVe. GloVe is an unsupervised algorithm built for natural-language corpora, repurposed here to treat a chronological medical history as a document.

The embedding model "encodes relationships amongst vector offsets representing the medical claims codes data" and reduces that space to coordinates. It then selects a minimum and maximum value for each dimension, across every code in the applicant's history (US Patent 12,682,400 B1, claim 1). In the illustrative embodiment GloVe maps codes into a 35-dimension Euclidean space, and the min-max aggregation across those 35 dimensions yields the 70 coordinates the random forest consumes.

The regression step is separate from the embedding step. A random forest is trained on those 70 coordinates plus age. It is fit against underwriting decision data for the roughly 40,000 applicants whose claims and outcomes MassMutual could match, from a dataset covering about 21,000 unique diagnostic codes and 8,000 unique procedural codes. The disclosed model form is close to austere.

Claim 1 also requires the system to generate an explanation tied to "the medical claims codes associated with the minimum values and the maximum values that resulted in a classification." Interpretability is claimed as part of the invention rather than bolted on. That does legal work too: a claim anchored in a traceable technical mechanism sits closer to the surviving side of the 2025-2026 Section 101 line for machine-learning claims.

A Claims-Derived Score Against a Priced Mortality Basis

The specification is not consistent about what the random forest predicts. The abstract describes training it "to predict relative mortality risk for underwriting applicants." The detailed description elsewhere describes the same model as trained to predict "underwriter-assigned risk rating." Those are different targets, and the training set, applications carrying "complete underwriting information such as final underwriting path and risk class," reads like the second.

That distinction reaches the pricing basis the moment the score leaves the underwriting desk. A carrier's priced mortality is calibrated to observed claim experience accumulated over years of exposure. A score trained to reproduce underwriter judgment inherits whatever the manual already encoded and offers no independent check on it, so it can accelerate applications faster without ever being validated against what the pricing basis measures.

The gap is sized. Accelerated underwriting programs run 15% average mortality slippage against priced assumptions, individual programs ranging from 5% to more than 30%. Carriers often lack the three to five years of seasoned claims experience needed to validate a program credibly. A patent claim can protect a pipeline. It cannot certify the output still means what the pricing actuary assumed a policy year later.

MassMutual already runs a fluidless program, MassExpress, that qualifies some applicants without labs or a paramedical exam, on underwriting rules, algorithms and a digital footprint (MassMutual, Underwriting Practices). A claims risk score is built for exactly that decision point, scoring an applicant directly instead of waiting weeks on an Attending Physician Statement.

Volume is what makes that compound. Munich Re Life US found the average acceleration rate with no human underwriter review at 11% as of late 2024, with carriers projecting 49% of total life business by 2030 (Munich Re, 2024). NAIC found 58% of 161 responding life insurers using, planning or exploring AI models, against 88% of auto insurers (NAIC, December 2023).

MassMutual has licensed mortality scoring since at least 2018, when LifeScore Labs partnered with Swiss Re on LifeScore360 (LifeHealthPro, March 2018). A five-patent family raises freedom-to-operate risk for reinsurer-supplied engines covering the same territory, and a carrier licensing a third-party module has a live question its vendor contract should answer.

Claim-Code Density as a Proxy

The density that makes the model interpretable is also what makes it a plausible proxy-discrimination vector. The number and specificity of ICD and CPT codes in a history is a function of how often a person saw a clinician and what was billed, not a direct measure of underlying health.

That is the mechanism Obermeyer and colleagues found driving racial bias in a widely used healthcare risk-prediction algorithm. It predicted cost rather than illness, and historically less is spent caring for Black patients at a given level of need, so it underestimated their risk. Correcting the bias would raise the share of Black patients flagged for extra care from 17.7% to 46.5% at the same clinical need (Obermeyer et al., Science, October 2019).

A claims-derived mortality score built on code density inherits a structurally similar risk. An applicant with less healthcare access generates a sparser, lower-dimensional claims signature that a random forest can read as lower risk, precisely backward from the truth for an underinsured population.

Regulators have written the testing obligation for exactly this. New York's Circular Letter No. 7 (2024) requires insurers using AI or external consumer data in underwriting to show the tools do not use, explicitly or as a proxy, any protected class status (NY DFS, July 2024). NAIC's Model Bulletin, in force in 23 states plus the District of Columbia, requires a documented governance program covering underwriting models (NAIC). The American Academy of Actuaries has warned that models inherit bias even when no protected-class variable enters directly (American Academy of Actuaries).

None of that is satisfied by the patent's explanation feature. Showing which embedding coordinates drove a classification tells a reviewer how the model reached its answer. It does not show whether claims-code density correlates with a protected characteristic through unequal access to care.