Ketoacidosis waits until the last C-peptide has gone

Design

secondary analysis of trial cohort data using crude and adjusted Andersen-Gill recurrent-event models with time-dependent covariates

Population

1441 participants with type 1 diabetes in the Diabetes Control and Complications Trial repository; 53% male, median age 27 years, average 6.5 years follow-up

Primary outcome

incidence of diabetic ketoacidosis by annually measured stimulated C-peptide above or below 0.2 nmol/mL

Effect

180 events in 129 participants; 179 (99.44%) after C-peptide 0.2 nmol/mL or below; hazard ratio for detectable C-peptide 0.07 (95% CI 0.01 to 0.48, P=0.007)

Residual beta cell function is usually discussed in terms of glycaemic control and hypoglycaemia. This analysis of the Diabetes Control and Complications Trial repository looked at ketoacidosis instead, using annually measured stimulated C-peptide and Andersen-Gill models for recurrent events with time-dependent covariates over an average 6.5 years of follow-up.

Among 1441 participants, 129 (9%) had 180 ketoacidosis events. The distribution is the finding: 179 of those 180 events - 99.44% - occurred after C-peptide had fallen to 0.2 nmol/mL or below in the preceding year. Only one occurred in someone with detectable C-peptide. The hazard ratio for detectable C-peptide was 0.07 (95% CI 0.01 to 0.48, P=0.007), and it held across adjusted models.

This is observational and it is an association, but the mechanism is straightforward and the separation is close to absolute. Two things follow. Clinically, an annual or periodic C-peptide in someone with long-standing type 1 diabetes is not just of academic interest - detectable levels mark someone at markedly lower ketoacidosis risk, and their disappearance marks a change in that risk. Strategically, it strengthens the argument for therapies that preserve endogenous insulin production, because the benefit would extend beyond glycaemic metrics to the acute complication that still kills people.

• Consider measuring C-peptide in long-standing type 1 diabetes; a detectable level marks much lower ketoacidosis risk.
• Read a fall to undetectable as a change in risk category, not just a laboratory curiosity.
• This is an association from trial cohort data, not a randomised comparison - do not present it as protection you can create.
• Intensify ketoacidosis education and sick-day planning most in patients with no residual C-peptide.
• Note the era: DCCT participants were managed before continuous monitoring and closed-loop systems, which may have shifted absolute risks.

The statistics, in plain English

A hazard ratio of 0.07 sounds implausibly strong until you see how it arises - one event in the entire detectable-C-peptide exposure time. That is also why the confidence interval is so wide, from 0.01 to 0.48: the estimate rests on a single event, and a single event cannot pin down an effect size. The direction is not in doubt; the magnitude is. Andersen-Gill models for recurrent events with time-dependent covariates are the right choice here, because they allow the same person to contribute both detectable and undetectable periods, which is what makes this more than a comparison between two groups of people.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.