Gairola 2022.

Study characteristics
Patient Sampling	Single‐centre, retrospective, case‐control study. First data set (64 participants, 114 eye samples) obtained from topography device, SmartKC at the Sankara Eye Hospital in Bengaluru, India. Second data set (2110 samples; 1637 non‐keratoconus and 473 keratoconus) consisted of downloaded anonymized data from the Keratron device database for all the people who took the corneal topography examination at the hospital from April 2008 to May 2010.
Patient characteristics and setting	First data set included 68 non‐keratoconus eyes and 46 keratoconus eyes. Second data set included 2110 samples, of which 1673 were non‐keratoconus and 473 were keratoconus.
Index tests	Convolutional neural network. The network is organized into 2 branches – 1 each for axial and tangential heatmaps – with a shared convolutional backbone, followed by 2 distinct feed‐forward classifiers, 1 for each branch. The shared backbone comprises the convolutional layers from a ResNet34 model. The model has a 2‐class (keratoconus versus non‐keratoconus) classification task. The article provides a clear explanation of the model and training procedure.
Target condition and reference standard(s)	The first data set was diagnosed by 1 senior ophthalmologist at the hospital. The diagnoses in the second data set were obtained based on the PPK‐based classification.
Flow and timing	All cases were included in the reference standard and index test. All data were included in a 2 × 2 table.
Comparative	Not applicable
Notes	No funding source mentioned.
Methodological quality
Item	Authors' judgement	Risk of bias	Applicability concerns
DOMAIN 1: Patient selection
Was a consecutive or random sample of patients enrolled?	Unclear
Was a case‐control design avoided?	No
Did the study avoid inappropriate exclusions?	No
Could the selection of patients have introduced bias?		High risk
Are there concerns that the included patients and setting do not match the review question?			High
DOMAIN 2: Index test (All tests)
Were the index test results interpreted without knowledge of the results of the reference standard?	Unclear
If a threshold was used, was it pre‐specified?	Unclear
Was the model designed in an appropriate manner?	Yes
Could the conduct or interpretation of the index test have introduced bias?		Unclear risk
Are there concerns that the index test, its conduct, or interpretation differ from the review question?			Low concern
DOMAIN 3: Reference standard
Is the reference standard likely to correctly classify the target condition?	No
Were the reference standard results interpreted without knowledge of the results of the index tests?	Yes
Could the reference standard, its conduct, or its interpretation have introduced bias?		High risk
Are there concerns that the target condition as defined by the reference standard does not match the question?			Low concern
DOMAIN 4: Flow and timing
Did all patients receive the same reference standard?	Yes
Were all patients included in the analysis?	Yes
Could the patient flow have introduced bias?		Low risk
DOMAIN 5: Comparative
Were different AI tests were developed and interpreted without knowledge of each other.
Are the proportions and reasons for missing data similar for all index tests?