Repeated sampling can reveal a biomarker's changing behavior. It can also inflate an apparent sample size if every specimen is mistakenly counted as a separate pregnancy. Primary study
Design and reported pattern
This prospective nested case-control study included 95 women with miscarriage contributing 173 samples and 265 women with healthy pregnancies contributing 557 samples. Kisspeptin's discrimination improved with gestational age, while the reported pattern for beta-hCG moved in the opposite direction. Combining markers improved the reported discrimination estimate. Primary report
Our appraisal: validation must respect repeated observations
The counts of people and specimens answer different questions. Precision calculations and validation procedures should recognize that measurements from the same pregnancy are correlated. Randomly splitting specimens rather than participants could allow information from one person to appear in both development and validation sets.
Gestational timing also belongs in the interpretation. A pooled performance estimate across the trimester can conceal stronger and weaker windows. A practical test needs a clear intended time of use, not simply an overall average.
Combining markers may add useful information, but it also creates a model that needs calibration and independent assessment. A decision matrix developed in one dataset should not be assumed to preserve its accuracy in a different population or assay setting.
Case-control sampling affects interpretation of absolute probabilities and overall accuracy. The proportion of cases in an analytic sample may differ from the frequency encountered in routine care. Sensitivity, specificity, predictive values, and calibration should therefore be kept distinct.
This report supports further biomarker validation, not individual diagnosis from a single number. It does not test administration of kisspeptin or show prevention of pregnancy loss. Release review should inspect participant-level analysis, timing relative to diagnosis, model development, and cohort overlap with related publications before interpreting it as independent confirmation.
Frequently asked questions
Were there more samples than participants?
Yes.
Are repeated samples independent pregnancies?
No.
Did performance vary with gestational age?
Yes.
Were combined-marker results reported?
Yes.
Does model accuracy automatically transfer to routine care?
No.
Was peptide treatment tested?
No.
What should a validation split keep together?
All relevant observations from the same participant or pregnancy.