Chastang 2000.

Study characteristics
Patient Sampling	Retrospective case‐control study, single‐centre (France) involving 208 corneal topographies (EyeSys System 2000) of 208 corneas from 8 groups of participants.
Patient characteristics and setting	Participants were classified by the following diagnoses. Normal Regular astigmatism Cataract surgery Radial keratotomy Excimer laser photorefractive keratectomy Non‐freeze myopic keratomileusis Penetrating keratoplasty Keratoconus
Index tests	Binary decision tree. In the first step, the distribution of keratoconic and non‐keratoconic patterns was studied based on the value of each index in the training set. For each index, corneas with an index value higher than the threshold (or cut‐off value) were classified as keratoconic corneas (positive test), whereas corneas with an index value less than the threshold were classified as non‐keratoconic (negative test). In the second step, binary decision trees were built by combining 2 indices to improve the classification method. The 6 indices with the highest sensitivity and specificity were used in these models. The first index was used to divide the training set into 2 populations (i.e. population with a positive test and population with a negative test) based on the previously calculated optimum threshold. In each of these populations, the distribution of keratoconic and non‐keratoconic patterns was studied based on the value of the second index. In each population, sensitivity and specificity curves as a function of the second index threshold were generated to evaluate the optimum cut‐off value. This resulted in 2 thresholds according to the response to the first test. In fact, the second index's most efficient threshold (i.e. the threshold with maximum sensitivity and specificity) in the population with a positive test was different from that in the population with a negative test. A cornea was classified as keratoconic when the second test was positive.
Target condition and reference standard(s)	Maps were classified by 2 cornea specialists based on clinical records and topographic appearances, before the index test.
Flow and timing	All cases were included in reference standard and index test. All data were included in a 2 × 2 table.
Comparative	Unclear whether different AI tests were developed and interpreted blind or independently and without knowledge of the results of each other, and whether missing data and their causes were similar for each AI test.
Notes	Supported in part by the Fondation Claude Bernard, Paris, France.
Methodological quality
Item	Authors' judgement	Risk of bias	Applicability concerns
DOMAIN 1: Patient selection
Was a consecutive or random sample of patients enrolled?	No
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?	Yes
If a threshold was used, was it pre‐specified?	Yes
Was the model designed in an appropriate manner?	Yes
Could the conduct or interpretation of the index test have introduced bias?		Low 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?	Yes
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?		Low 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.	Unclear
Are the proportions and reasons for missing data similar for all index tests?	Unclear
		Unclear risk