When a Death Becomes a Price Signal in Ghana
A policy analysis in npj Health Systems asks whether facility mortality data could help recalibrate Ghana's DRG tariffs — then spends most of its length on why the data, and the incentives around it, are not ready for that job.
Ghana has paid its hospitals by diagnosis since 2008. Under the Ghana-Diagnosis Related Groupings system an admission is sorted into one of more than 550 clinical groupings and the National Health Insurance Scheme pays a fixed tariff for it rather than reimbursing what the ward spent. The reform was meant to end the incentives of open-ended fee-for-service, and it has brought its own difficulties: tariffs that drift from real costs, and upcoding, in which a case is filed into a better-paid group than it belongs to.
A policy analysis published on 4 November in npj Health Systems asks whether a different kind of data could be folded into that machinery — not what the hospital spent, but who died. Regina Titi-Ofei and colleagues, from the University of Geneva, the Global Fund, the WHO Regional Office for Africa and Ghanaian research institutions, set out how facility mortality records might inform the revision of groupings and tariffs. The paper generated and analysed no datasets of its own. It is an argument, and an unusually self-limiting one, much of it spent on what would have to be true before it could be tried.
The distinction the authors keep insisting on
The first thing the paper does is block a misreading. Mortality, it says, should not set reimbursement. What it proposes is narrower: in-facility deaths and case fatality rates used as a diagnostic on the classification itself, to find groups that are internally heterogeneous, where one tariff covers patients with very different risks of dying. Persistent excess mortality inside a grouping would suggest it masks severity differences that mature systems absorb through risk adjustment built into the payment formula; a WHO review of DRG schemes in low- and middle-income countries describes what happens where patient-level costing is missing — simplified groupings and negotiated tariffs. On this account, a death signal is a cheap stand-in for the cost accounting Ghana does not have.
A second function is kept separate: risk-adjusted mortality as a quality indicator, pointing at referral pathways rather than case mix. And the authors say what must not be built — “higher observed mortality should not result in higher reimbursement”, because that would invert the incentive a hospital faces.
Three numbers the perspective does not carry
The logic is coherent. Whether it can be executed depends on the input, and the input has been measured elsewhere.
Agyei-Asabere and colleagues, in a PLOS ONE study published in July, evaluated Ghana’s 2022 mortality register against two censuses using two demographic techniques. Completeness came out at 28.7% under the empirical completeness method and 56% under death distribution methods — a factor-of-two disagreement between the methods themselves — and subnationally at nearly 70% in Greater Accra against barely 8% in the Oti Region, with under-five deaths noticeably under-reported and little improvement over two decades in a system that dates to 1888.
Owusu and colleagues, reading Ghana Health Service district returns for 2014 to 2018, found institutional mortality falling 7% nationally while rising in four of ten regions, by 17% in the Upper East and 13% in the Volta. Cause of death existed for only the last two years of the series and, grouped to ICD-11, showed non-communicable diseases leading at 25% then 20%, infections at 15% in both, and hypertension largest with 2,243 and 2,472 deaths. That is a descriptive read of routine returns, not clinical certification.
The third number belongs to the price side. Gyasi and colleagues compared G-DRG tariffs with approved charges at the Greater Accra Regional Hospital for 2018 to 2022, across the ten largest inpatient conditions. Tariffs rose in steps of about 16% every other year; charges moved erratically, in increments of 20% to 58%. Over five years the differential left the facility a cumulative 23% gain — and the biggest block of that volume was paediatric: 3,323 admissions in 2022, 63.5% of the top ten conditions’ total, were child-related; neonatal jaundice alone was 1,177, or 23%. Tariff design here sits on top of newborn care.
The deaths outside the gate
The bias the paper names is one this digest returns to often. Facility deaths are a subsample — sicker than average, and drawn only from those who reached care. The authors list the standard remedies, redistributing ill-defined causes and calibrating against community data, and propose triangulating any mortality signal against sources outside the hospital, including Ghana’s surveillance sites at Navrongo, Kintampo and Dodowa.
Those sites are where population-level cause-of-death measurement in Ghana actually happens, and their cause assignment runs through verbal autopsy: structured interviews with bereaved families, read by physicians or by probabilistic models such as the InterVA model, for deaths that never produced a medical certificate. The work is not free — a costing of a surveillance programme in Andhra Pradesh put it at roughly sixteen dollars per death in its first year, falling to about twelve — and a financing reform built on hospital records will not pay for it. The payment-relevant subset is smaller than the registered subset, which is itself somewhere between 28.7% and 56% of all deaths. A severity signal calibrated on that minority describes the patients who arrived.
An incentive that runs the other way
Here the argument stops being about accounting. If mortality patterns feed into how groupings and tariffs are revised, the cause-of-death field changes character: it stops being a statistical residue and becomes a financial one. Ghana has already seen claims inflation under the G-DRG, and the paper lists upcoding among its documented problems, warning that a badly designed mortality link could intensify the behaviour — misreported diagnoses, exaggerated severity, patients sorted into better-paid groups. The temptation falls hardest on the fields hardest to verify afterwards.
Certification is already unreliable when nobody is watching. A national audit of 25,123 Iranian death certificates, in the WHO Eastern Mediterranean Health Journal and covered here in October, judged 29.3% correctly completed and found the interval between onset and death missing in 86.3% of cases. Give the same form an audience with money attached, and the error rate stops being an accident.
What the authors ask for before any of it
The prescriptions are procedural rather than numerical. Pilot in maternal, neonatal and selected infectious conditions, where mortality recording is comparatively complete. Triangulate against surveillance sites, civil registration and WHO and Global Burden of Disease estimates rather than trusting facility returns. Couple any payment change to ICD-11 training, clinical documentation and routine mortality audits, with independent coding audits and transparent risk adjustment tied to the electronic claims platform. Consult widely: Ghana’s earlier capitation pilot in the Ashanti Region foundered partly on trust.
One risk deserves more weight than its single paragraph. Hospitals with better digital infrastructure and better coders would document severity more accurately, and would benefit more from severity-linked payment. A data-quality advantage becomes a funding advantage, and rural facilities serving the poorest patients are penalised for their records rather than their care.
The perspective’s contribution is to move an old argument onto new ground. For two decades the case for better cause-of-death data has rested on invisibility: uncounted deaths produce bad policy, and the instruments that count them cost money. This paper adds the other half — that once a death is counted, priced and paid for, it acquires a reason to be recorded inaccurately. Both effects land in the same column of the same table, and only an audit separates them.
Sources
- Titi-Ofei R, Osei Afriyie D, Kipruto H, Nonvignon J, Rossier C, Adofo-Ofosu A. Opportunities, risks and policy implications of using health facility mortality data to inform DRG payment design under Ghana's NHIS. npj Health Systems 2026;3:105, published 4 November 2026 (policy analysis; no datasets generated or analysed) pmc.ncbi.nlm.nih.gov
- Agyei-Asabere C, Bawah AA, Frempong-Ainguah F, Baumgartner J, Agyei-Mensah S, Botchway E, Abordo VS, Ezzati M. Completeness of death registration in Ghana: An evaluation of multiple data sources and methods. PLOS ONE 2026;21(7):e0354361, published 23 July 2026 pmc.ncbi.nlm.nih.gov
- Owusu AY, et al. Institutional mortality rate and cause of death at health facilities in Ghana between 2014 and 2018. PLOS ONE 2021;16(6):e0256515 pmc.ncbi.nlm.nih.gov
- Gyasi P, Darkwa GC, Aryeetey E, Nsiah-Boateng E. Ghana's Diagnosis Related Grouping (G-DRG) tariffs versus hospital charges: implications for inpatient care financing at a secondary hospital in Ghana. Cost Effectiveness and Resource Allocation 2026;24, published 22 May 2026 pmc.ncbi.nlm.nih.gov
- Mathauer I, Wittenbecher F. DRG-based payment systems in low- and middle-income countries: implementation experiences and challenges. World Health Organization (WHO IRIS, handle 10665/75150) iris.who.int
- Joshi R, et al. Cost and cost-effectiveness of a verbal autopsy programme: the Andhra Pradesh experience. PLOS ONE 2015;10(5):e0126410 pmc.ncbi.nlm.nih.gov
- Kazemeini H, Khosravi A, Atay A, Rashidian E, Rabbani B, Kazemi E, et al. Analysis of the quality of death certification in Islamic Republic of Iran. Eastern Mediterranean Health Journal 2026;32(1):37-43 (WHO IRIS, handle 10665/385020) iris.who.int