摘要
A minority of patients with diabetic ketoacidosis (DKA) remain profoundly acidaemic, despite receiving standard treatment as per established guidelines. This review examines the reported use of renal replacement therapy in that setting, proposes a working definition of refractory DKA, and sets out the differential diagnosis of persistent acidaemia that should be excluded before escalation is considered. Targeted, purposive, non-systematic search of PubMed with citation tracking of retrieved articles. No protocol was registered, and no formal screening or risk-of-bias assessment was undertaken. The retrieval was not intended to be exhaustive, and completeness should not be inferred. To date, no clinical trial of bicarbonate has included patients with a pH below 6.85, although the current consensus permits its consideration below pH 7.0. Expert guidance on extra-renal purification exists for metformin-associated lactic acidosis, toxic alcohol poisoning, and salicylate toxicity; however, none addresses DKA. Reported cases document escalation for persistent acidaemia, yet circuit composition, anticoagulation, and insulin handling are infrequently recorded, and insulin clearance across a haemofilter has not been measured in this setting. Continuous therapy using glucose-free solutions has itself precipitated euglycaemic ketosis, including in people without diabetes. Ketoacidosis associated with sodium–glucose co-transporter 2 inhibition often presents below the glucose threshold used here and requires separate consideration, although escalation is more often described in that setting, including one report in which acidaemia persisted. A working definition of refractory DKA is proposed for further evaluation. Persistent acidaemia often has an alternative cause that should be excluded first, but, where that exclusion is complete and acidaemia persists, renal replacement therapy has been used and reported, and weak expert guidance supports it in analogous acidoses; no benefit has been shown in DKA. Comparative evidence is lacking, and individual patient data synthesis is the next necessary step.