Ketone bodies are a normal fuel. When carbohydrate intake is low and insulin concentrations fall, adipose tissue releases fatty acids, the liver converts part of them to acetoacetate and β-hydroxybutyrate, and the brain and muscle oxidise these in place of glucose (Laffel, 1999). In prolonged fasting, ketone concentrations rise over days to a plateau that is held in check by the small amount of insulin still secreted, and blood pH stays within the normal range (Cahill, 2006). The ketosis of a well-conducted very low carbohydrate diet is of this regulated kind. Ketoacidosis differs in degree and in control: insulin action is insufficient to restrain lipolysis and ketogenesis, counter-regulatory hormones are raised, ketone production outruns use, and buffering capacity is exhausted (Dhatariya et al., 2020). The distinction is made biochemically, by acid-base status, and not by the presence of ketones alone.

Glucose stays low in ketoacidosis when its supply is reduced or its loss increased. Reduced supply occurs with fasting, low carbohydrate intake, depleted hepatic glycogen, alcohol and pregnancy. Increased loss is the mechanism of SGLT2 inhibition, which maintains glycosuria independent of insulin. By lowering glucose these drugs also lower insulin secretion and raise glucagon, shifting metabolism toward ketogenesis, and they may reduce renal clearance of ketones (Taylor et al., 2015). Rosenstock and Ferrannini (2015) describe the result as predictable and list reduced carbohydrate intake, acute illness, surgery, dehydration, alcohol and reduction of insulin doses among the usual precipitants. A very low carbohydrate diet in a person taking an SGLT2 inhibitor therefore combines two influences that act in the same direction, and separating them in a given case requires careful documentation.

Case reports are the usual first signal of a rare harm and have a legitimate place in the literature (Vandenbroucke, 2001). They cannot give a rate, and causal inference from a single episode is always provisional. Pharmacovigilance has developed tools for making that inference more disciplined, including the questionnaire of Naranjo et al. (1981) and the categorical scheme described by Edwards and Aronson (2000). Reporting standards have developed in parallel: the CARE guidelines specify what a case report should contain (Gagnier et al., 2014; Riley et al., 2017), and a short tool exists for appraising the quality of case reports and series (Murad et al., 2018). None of these was written with a dietary exposure in mind.

The report should show that the episode met accepted criteria. The current consensus requires evidence of diabetes or of hyperglycaemia, ketonaemia with β-hydroxybutyrate of 3.0 mmol/L or more or marked ketonuria, and metabolic acidosis with pH below 7.3 or bicarbonate below 18 mmol/L (Umpierrez et al., 2024). Earlier criteria set the glucose threshold at 250 mg/dL (Kitabchi et al., 2009), and the term euglycaemic is conventionally applied below that value. Blood β-hydroxybutyrate is preferable to urine testing, which detects acetoacetate. The values at presentation, the method of ketone measurement and the anion gap should all be stated.

Diet is the exposure of interest and is usually the least well described element. Definitions vary; one widely cited scheme describes a very low carbohydrate ketogenic diet as 20 to 50 g of carbohydrate a day, or less than 10 per cent of energy (Feinman et al., 2015). The report should give the estimated daily carbohydrate intake and how it was estimated, total energy intake if known, the duration of the diet, whether it was supervised, and any recent change such as a further restriction, a period of fasting or reduced intake during illness. Recent weight loss and any home ketone readings before the episode are also informative.

Every glucose-lowering medicine should be listed with its dose and the time of the last dose, with particular attention to SGLT2 inhibitors and to recent reduction or omission of insulin. Because insulin deficiency is the central lesion of ketoacidosis, the report should state what is known of endogenous secretion. A proportion of adults labelled as having type 2 diabetes have slowly progressive autoimmune diabetes or pancreatic disease, and a measurement of C-peptide and islet autoantibodies, taken once the acute episode has resolved, helps to show whether the diagnosis was secure.

The recognised triggers of ketoacidosis should be sought and their presence or absence stated: infection, myocardial infarction and other acute illness, surgery, vomiting and dehydration, alcohol, pregnancy, and strenuous exertion (Kitabchi et al., 2009; Rosenstock and Ferrannini, 2015). A statement that no precipitant was identified carries weight only if the investigations that support it are named.

A raised anion gap with ketones is not unique to diabetes. Alcoholic ketoacidosis, starvation ketosis with a superimposed acidosis, lactic acidosis, renal failure and ingestion of salicylate, methanol or ethylene glycol belong in the differential diagnosis (Dhatariya et al., 2020). The report should give lactate, renal function and the alcohol history at least, and say which further tests were done.

Causality tools for adverse drug reactions ask a small number of questions that transfer readily to a dietary exposure: whether the event followed the exposure in a plausible time, whether it is a recognised or mechanistically plausible effect, whether other causes could explain it alone, whether it resolved when the exposure was changed, and whether it recurred on re-exposure (Naranjo et al., 1981). The framework proposes that authors answer each question explicitly for the diet and separately for any SGLT2 inhibitor, and then place the diet in one of the familiar categories of certain, probable, possible or unlikely (Edwards and Aronson, 2000). Where an SGLT2 inhibitor was being taken, or another precipitant was present, the diet can rarely be rated above possible or contributory, and the report should say so. Deliberate re-exposure to establish cause is not justified; information on what happened when carbohydrate restriction was later resumed, if the person chose to do so, should be reported when available. The general considerations set out by Hill (1965), particularly temporality and plausibility, provide a useful check on the reasoning.

The write-up should follow the CARE checklist, with a dated timeline that places dietary changes, medicine changes, symptoms and laboratory results on one axis (Riley et al., 2017). Informed consent, the perspective of the person concerned where they wish to give it, and the plan for diet and medicines after the episode complete the report. Negative findings belong in the text; their omission is the commonest reason a case report cannot be interpreted.