FGF23 in acute kidney injury: moderate at best, and the assay matters

Design

systematic review and meta-analysis of diagnostic and prognostic accuracy, random-effects bivariate model, QUADAS-2 quality assessment

Population

23 studies of circulating intact or C-terminal FGF23 in acute kidney injury; 17 diagnostic, 6 prognostic

Primary outcome

pooled sensitivity, specificity and summary ROC area for diagnosis of acute kidney injury and for mortality

Effect

diagnosis sensitivity 0.79 (95% CI 0.73 to 0.86), specificity 0.82 (0.75 to 0.89), SROC AUC 0.87 (0.81 to 0.92), I2 67%; mortality SROC AUC 0.77 (0.70 to 0.83)

A systematic review and meta-analysis pooled 23 studies of circulating fibroblast growth factor 23 in acute kidney injury - 17 diagnostic, 6 prognostic - covering both intact FGF23 and the C-terminal assay, using a random-effects bivariate model.

For diagnosis, pooled sensitivity was 0.79 (95% CI 0.73 to 0.86) and specificity 0.82 (0.75 to 0.89), giving a positive likelihood ratio of 4.40, a negative likelihood ratio of 0.25, a diagnostic odds ratio of 17.49 and a summary ROC area of 0.87 (0.81 to 0.92). The intact assay clearly outperformed the C-terminal one, with areas of 0.91 against 0.81. For predicting mortality after acute kidney injury the performance was weaker: sensitivity 0.77, specificity 0.76, summary area 0.77 (0.70 to 0.83). Heterogeneity was substantial throughout - I squared 67% for diagnosis and above 80% for the prognostic estimates. No publication bias was detected.

The authors' own conclusion is the right one: this is a complementary biomarker for risk stratification, pending validation with standardised protocols. A likelihood ratio of 4.40 shifts a pre-test probability meaningfully but does not decide anything on its own, and a negative likelihood ratio of 0.25 does not rule acute kidney injury out. Nothing here justifies ordering FGF23 in routine practice, and the assay heterogeneity means a result from one laboratory is not interchangeable with one from another.

• Do not order FGF23 to diagnose acute kidney injury outside a research protocol - the evidence does not support it as a standalone test.
• If a study or laboratory offers it, establish which assay is being used; intact and C-terminal FGF23 performed differently.
• Read a likelihood ratio of 4.40 as a moderate shift, not a rule-in - it will not replace serial creatinine and urine output.
• Treat the mortality prediction as weaker still: a summary area of 0.77 is modest for a prognostic marker.
• Watch for standardisation before this becomes usable; the heterogeneity here is the obstacle, not the effect size.

The statistics, in plain English

I squared of 67% for the diagnostic estimates, and above 80% for the prognostic ones, means most of the variation between studies is real difference rather than chance - so the pooled sensitivity and specificity describe an average across settings that may not apply to yours. A summary ROC area of 0.87 sounds respectable, but the likelihood ratios are the practical translation: 4.40 for a positive result moves a 20% pre-test probability to about 52%, and 0.25 for a negative moves it to about 6%. Neither is decisive. The gap between the intact and C-terminal assays, 0.91 against 0.81, is larger than the gap between a useful test and a marginal one, which is why assay type is not a technical detail here.

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.