The 3% False Negative: Deploying a Sepsis Algorithm
A public hospital's ethics committee has 12 members and must advise on a sepsis-prediction algorithm. The hospital admits 45,000 patients a year and treats about 1,200 sepsis cases. The vendor reports a 3% false-negative rate among sepsis cases and an 8% false-positive rate among all admissions; the tool improves detection by 18% compared with current practice. It also increases alert workload and stores patient data for 24 months. Two proposals are on the table: full deployment with an opt-out for patients, or a six-month pilot in two wards with mandatory human override. Clinicians worry about alert fatigue; patient representatives demand meaningful consent; the finance officer notes that the vendor's contract is already signed. The committee must vote at its next meeting and cannot commission a new external trial.
1. Identify the morally relevant features of the decision and explain why the 3% false-negative rate cannot be assessed in isolation.
Read the worked answer
Morally relevant features include: the expected number of missed sepsis cases, the number of false alerts and their effect on clinical attention, the severity of harm from a missed case, patient autonomy and consent, distribution of benefits and burdens across wards and patient groups, data retention and privacy, and the reliability of human override. The 3% figure is not self-interpreting. Three per cent of 1,200 sepsis cases is about 36 missed cases, but 8% false positives across 45,000 admissions is about 3,600 alerts; those alerts consume staff time and can cause harm through delay and desensitisation. The 18% detection improvement must therefore be weighed against false-positive burden, workflow effects, and the possibility that the tool performs unevenly across subgroups. A bare false-negative rate cannot settle the ethics.
2. Compare a Kantian and a utilitarian assessment of the opt-out deployment proposal.
Read the worked answer
A Kantian assessment centres on autonomy and respect for persons. Opt-out consent can be legitimate only if patients have real, informed, easy choice and are not merely used as means to population health. If the notice is buried and opting out is burdensome, deployment treats patients as instruments. The duty of care also cuts the other way: clinicians may have an obligation to use a tool that reduces missed sepsis. A utilitarian assessment centres on expected net welfare: 18% better detection may mean fewer deaths and less suffering, while 3,600 false alerts produce anxiety, cost, and delay. A rule-utilitarian would also ask whether opt-out deployment erodes trust in the hospital. The two frameworks may converge on a cautious pilot with strong consent, but for different reasons: Kant stresses respect, utilitarianism stresses net outcomes and institutional trust.
3. Recommend a course of action using at least two frameworks and specify safeguards. Address the moral remainder.
Read the worked answer
Recommendation: adopt a six-month pilot in two wards, not full opt-out deployment. From principlism, this satisfies beneficence and nonmaleficence while protecting autonomy through explicit notice and easy opt-out; from a utilitarian standpoint, it generates local evidence on alert burden and subgroup performance before scaling. Safeguards: mandatory human override, clinician training, audit of false negatives and false positives by ward, independent review, data minimisation and a shorter retention period unless justified, and a clear route for patients to decline without losing ordinary care. Moral remainder: even a good pilot will miss some sepsis cases and will expose some patients to privacy intrusion. Those harms are not cancelled by the net benefit; they require monitoring, compensation where appropriate, and a commitment to stop or redesign if the pilot reproduces inequitable outcomes.