What is a medication reconciliation catch worth?

A dialysis patient is found, during a monthly reconciliation, to be on a drug that is misdosed, contraindicated, or otherwise inappropriate for ESKD. The drug is stopped or corrected. What was that worth?

The honest answer has three parts, and only the first is a dollar figure:

1. Expected harm avoided, monetizable where dialysis-specific attributable risk exists.
2. Process work demonstrably done, countable but not monetizable, where the drug is a hard stop on mechanism or label grounds but no incidence data exists.
3. Uncertainty surfaced, where the right action is clinician review and claiming value in either direction would be indefensible.

Most published MedRec value models collapse all three into one number. That is why they do not survive payer review. This framework keeps them separate.

1. The core identity

The central move is converting an exposure metric into a probability of harm metric. Counting days off an inappropriate drug is the right raw material, but days are an input, not a value.

Expected events avoided  =  Baseline event rate (per patient-year)
                          x Attributable fraction among the exposed (AF_e)
                          x Exposure-time averted (patient-years)

Value  =  Expected events avoided  x  Cost per event

                   AF_e  =  (HR - 1) / HR

AF_e is the fraction of events occurring in exposed patients that are attributable to the exposure. It is the bridge between "this patient was on gabapentin 600 mg daily for 90 days" and "we avoided 0.012 falls." A hazard ratio of 1.55 means 35.5% of falls in exposed patients are attributable to the exposure, so removing it removes 35.5% of that patient's fall risk, not all of it.

Why not just count days

• Risk is dose-dependent and front-loaded. Gabapentin above 300 mg/day carries HR 1.55 for falls; at 100 to 200 mg/day it is 1.35. Days are not fungible across doses.
• The counterfactual is unbounded. "Days off the drug" implies the patient would otherwise have stayed on it forever. Anchor the horizon to observed refill behavior (PDC) or a defensible censoring rule, or the model inflates without limit.
• Days are not an outcome anyone pays for. Events and admissions are.

The counterfactual horizon

Exposure-time averted must be bounded by something observable. In descending order of defensibility: observed refill and PDC data for that patient; population median time-on-drug for the agent; or a fixed conservative horizon, censored at the next scheduled reconciliation.

Note the interaction with a monthly cadence. If reconciliation happens every month, the honest counterfactual for most findings is not indefinite exposure, it is exposure until the next review. A monthly program's per-catch value is therefore lower than an annual program's, but its total value is higher because it catches more. Modeling monthly cadence with an annual exposure horizon double-counts the program's own benefit.

2. The evidence tiers

Not every inappropriate medication can be assigned a dollar value, and pretending otherwise is where these models lose credibility. Each drug is tiered by evidence quality, not clinical conviction.

A drug being in Tier C says nothing about whether it should be stopped. Metformin at ESKD is an absolute contraindication and a hard stop, and it is Tier C, because no one has published the incidence of metformin-associated lactic acidosis per exposure-year in dialysis. Clinical certainty and economic quantifiability are different axes. Conflating them is the most common failure in this literature.

3. Cost-per-event anchors

All figures inflated to 2026-M07 USD using CPI-U Medical Care (BLS series CUUR0000SAM, index 593.781).

Pellegrin is the primary anchor because it is US, Medicare, claims-auditable, and comes from the same paper that supplies the ICD code set defining the denominator, so numerator and cost are methodologically coupled. Its limitation is that it is an age-65+ general population, not dialysis.

Dollar-year warning. The Bates and Poudel dollar years are not stated in their sources; the years above are assumptions. For Bates this matters: a roughly 2.8x inflation factor over 32 years makes that anchor fragile.

One anchor deliberately excluded

A widely circulated figure puts an adverse-drug-reaction admission at EUR 9,538 (Bennett 2023). That should not be used as an avoided-cost parameter. It is the gross admission cost for ADR patients; the non-ADR comparator in the same study cost more (EUR 9,828). The paper's attributable estimate is EUR 2,047 (95% CI -889 to 4,983), not significant, with every severity and preventability interval crossing zero. Using EUR 9,538 as an avoidable cost inverts the paper's finding. It is excluded here, with the reason recorded in code so it is not silently reintroduced.

4. What this model does not claim

• Endpoints overlap and must not be summed. Gabapentin's altered-mental-status and fall endpoints come from the same cohort and the same patients. Only one endpoint per drug is modeled.
• Anticoagulant comparators are not "no drug." Dabigatran's RR 1.78 is versus warfarin. The counterfactual for a catch is a switch to warfarin or apixaban, so the value is the difference in risk, not the whole risk.
• Mortality endpoints need a different model. Digoxin's HR 1.28 is for death. Multiplying it by a hospitalization cost is a category error.
• The baseline-rate approximation is deliberately conservative. Published cohort-wide rates mix exposed and unexposed patients; using them understates value by roughly 14 to 27% for the drugs modeled. The calculator exposes both settings.
• Program effect sizes are not additive with drug-specific ones. Meta-analytic MTM odds ratios and drug-specific attributable-fraction calculations are two routes to the same quantity. Stacking them double-counts.

5. The recommended composite metric

Report five things. Only the last is a dollar figure, and it is reported as a range.

1. Medication-related hospitalization rate per 1,000 patient-months, ICD-defined per the CMS/AHRQ code set. Auditable from claims.
2. High-risk-drug dosing-appropriateness rate: the share of patients on renally-adjusted or discontinued gabapentinoids, opioids, DOACs, and the Tier C hard stops.
3. Drug-related problems identified and resolved per patient. Reconciliation increases discrepancy resolution roughly sevenfold (RR 7.48, 95% CI 5.62 to 9.95, moderate certainty), though from a single RCT of 564 patients.
4. Inappropriate-exposure-days averted, with the counterfactual horizon stated.
5. Modeled cost avoided, as a range using at least two published unit costs, with Tier C and D findings reported alongside as counts, not dollars.

6. Honest limitations

• No dialysis-specific cost-per-event anchor exists. Every cost figure here is from a general or age-65+ population. Dialysis patients are hospitalized more often and at higher cost, so the anchors are probably conservative, but "probably conservative" is an assumption, not a finding.
• The strongest systematic review of medication therapy management graded the economic evidence low to insufficient for most outcomes, including drug therapy problems, adverse drug events, and mortality. These are modeled estimates, not measured savings.
• The most-cited ROI figure in this space ($18.50 per $1) is an outlier: a single health plan, 95% CI $9.10 to $27.70, against a $1.29 comparator from an integrated system. Use it as a sanity ceiling, not a benchmark.
• Several drugs have no usable quantification in dialysis: colchicine baseline use (the only controlled ESKD study, n=22, was null, and is a survivor cohort), NSAIDs (the one large dialysis dataset found a null association), enoxaparin, famotidine, and statin interactions.
• Deprescribing has costs of its own. DOPPS documented substantial under-treatment of pain in dialysis: three-quarters of patients reporting moderate-to-severe pain had no analgesic prescribed. A program that removes gabapentinoids and opioids without a replacement analgesic plan is not obviously improving care, and a model counting only avoided falls will not see that.

Sources

Effect sizes: Ishida JASN 2018;29:1970-78 and CJASN 2018;13:746-53; Chan Circulation 2015;131:972-9 and JASN 2010;21:1550-9; Muanda JAMA 2019;322:1987-95; Siontis Circulation 2018;138:1519-29; Hung CID 2005;41:291-300; Blumenberg J Med Toxicol 2020;16:222-29; Joglar JACC 2024;83:109-279.

Cost and program effects: Pellegrin JAGS 2017;65:212-19; Bates JAMA 1997;277:307-11; Poudel PDS 2017;26:635-41; Deng Front Pharmacol 2023;14:1143444; Bulow Cochrane 2023;1:CD008986; Ciapponi Cochrane 2021;11:CD009985; Viswanathan JAMA Intern Med 2015;175:76-87; Peasah JMCP 2021;27:147-56.

Inflation: US Bureau of Labor Statistics, CPI-U Medical Care, series CUUR0000SAM, retrieved 2026-09-10.