This institutional study was set at the Department of Ophthalmology, Semmelweis University, Budapest, Hungary. In this study, 44 eyes of 24 patients (15 men, 9 women) from keratoconus suspect cases to moderate keratoconus cases were included (age: 35.25 ± 10.67). Severe cases were excluded, because of potential stromal haze or scar formation, which may alter the optical transparency of the cornea and image acquisition of the Pentacam. Both eyes of each patient have undergone a complete ophthalmologic evaluation including slit-lamp biomicroscopy, keratometry, retinoscopy, ophthalmoscopy and Placido disc-based videokeratography. Keratoconus suspect cases were diagnosed when topographic abnormality was observed according to the Rabinowitz criteria without any slit-lamp findings (Rabinowitz 1995; Rabinowitz & Rasheed 1999). The criteria for diagnosing keratoconus were defined as the existence of central thinning of the cornea with Fleischer ring, Vogt’s striae or both by slit-lamp examination in addition to topographic findings. In the control group, 70 eyes of 41 refractive surgery candidates (24 men, 17 women) were included (age: 39.69 ± 15.77). Both eyes of each subject were used in the study, except for 12 fellow eyes because of previous refractive surgery or trauma. Control subjects were age-matched. Patients, who wore rigid contact lenses, were asked to stop using them for 4 weeks, and soft contact lenses were ceased for at least 1 week before assessment. All eyes were examined with the Pentacam HR (version number: 1.16 r: 23), used by three trained examiners without application of dilating or anaesthetic eye drops. The readings were taken as recommended in the instruction manual. Briefly, the patients were instructed to keep both eyes open and fixate on the black target, in the centre of the blue fixation beam. After attaining perfect alignment, the instrument automatically took 25 Scheimpflug images within 2 seconds. The measurement results were checked under the quality specification (QS) window, only the correct measurements (‘QS’ reads OK) were accepted; if the comments were marked yellow or red, the examination was repeated. For local posterior elevation measurements, four different reference surfaces were chosen: the BFS with autodiameter and fixed 8-mm-diameter settings and the best fit toric ellipsoid (BFTE) with autodiameter and fixed 8-mm-diameter settings. Compared to values of posterior elevation by BFS with fixed 8 mm diameter, those obtained by 9 mm showed an increased variance, indicating that further analysis would be less reliable (data not shown). For all methods, the float map was chosen, which means that the reference body has no fixed centre, and the average distance between posterior surface of the cornea and the sphere surface is optimized to be as small as possible. Elevation maps show the difference in height between cornea and reference body; their value is positive, when the measured point of the cornea is above the reference body, and negative, when it lies below. Posterior elevation data were read at the thinnest point of the cornea. Posterior corneal surface asphericity (Q value) was calculated at the sagittal angle ring at 30° centred on the apex. Statistical analyses were performed with SPSS 15.0 software (SPSS Inc., Chicago, Il, USA). Normal distribution assumption could be accepted for all of the variables according to the Shapiro–Wilks’ W test. For group comparisons of continuous variables, the independent sample t-test was used. A p value < 0.05 was considered statistically significant. Because binocular data of patients were evaluated in this study, we proposed the use of the bootstrap resampling technique as a tool to assess within-subject correlations. The basic idea of the bootstrap (Efron 1979) is to produce a random sample, which is obtained by sampling, with replacement, from the original pool of data. The bootstrap sample is then used to compute the estimate of the parameter, and this procedure (extraction of the random sample and computation of the estimate) is repeated 3000 times to create a nonparametric distribution of the parameters providing the estimated mean and the 2.5% and 97.5% confidence limits (95% CI). Receiver operator characteristic curves (ROCs) were used to compare discriminating ability and to determine cut-off values of posterior elevation measurements obtained by different reference surfaces. Intraclass (between eyes of the same patient) correlation was overcome using booststrap method deriving estimates of mean and confidence intervals for area under the ROC (AUROC). Comparison of AUROC values of different indices was made to test significant differences between reference surfaces. To explore the effect of corneal cylinder and asphericity on different calculation methods of posterior elevation, four independent generalized estimating equation (GEE) models were constructed using posterior elevation values calculated by the four reference surfaces as continuous dependent variables. GEE models take into account between-eye correlations by treating data from eyes of patients in statistical analysis as repeated measures (Zeger et al. 1988; Hanley et al. 2003). GEE is valid with data missing completely at random, i.e. analysing some patients with data of one eye only (Touloumi et al. 2001) as occurred in many previous studies. Comparisons of different regression models for a given dependent variable were assessed by the value of the Corrected Quasi Likelihood under Independence Model Criterion (QICC). Lower value of QICC indicates better fit to data (Zeger et al. 1988; Hanley et al. 2003). Finally, Bland–Altman limits of agreement were calculated to determine the agreement between the most different reference surfaces across posterior elevation values.