The primary method to identify keratoconus progression is to monitor for changes in corneal shape, particularly an increase in keratometry. However, we have shown that with the same dataset, different parameters, used singly or in combination, give widely different estimates. The reason for this is uncertain. It is known that the accuracy of keratometry reduces as the cornea becomes more ectatic, and that using one threshold will not identify changes in keratometry with the same accuracy over the whole range of keratoconus severity.15–17, 23 As a result, with one threshold, there is a tendency to over-diagnose progression in steep corneas, and miss a repeatable increase in keratometry in flat corneas. A partial solution to this problem is to use a higher threshold for more advanced disease.14,15 We have refined this approach by identifying the regression-based 95% limits of agreement (LoA) for keratometry to derive thresholds that can be applied individually to all values of corneal curvature. We have then used these adaptive thresholds to quantify keratometry changes in a population following CXL and then compared these rate of progression with values obtained with standard methods and thresholds.7,10,11,24,25 We have also incorporated a delayed time point for base-line observations to allow any acute changes of corneal shape following CXL to stabilise.6, 18–20
Our results show that keratoconus can continue to progress after CXL for at least five years following the baseline, but with a wide range (6–29%) in the estimates of progression depending on the parameter used. For example, with a standard fixed front K2 threshold of ≥ 1.5D, or the multi-parameter method, the number of eyes that had progressed five years after the reference visit was similar at 6% and 8%, respectively. However, with a standard fixed parameter Kmax threshold of ≥ 1.0D, the progression rate was much higher at 29%, although this reduced to 18% if we increased this threshold to ≥ 1.5D. In contrast, the results using the three adaptive thresholds were more uniform at 20–22%. Importantly, because there is no reference standard (ground truth) to define progression, we can only compare the results obtained with the different methods, but we do not know which is the most accurate or which best represents progression. It is also difficult to compare our estimates with other case series due to the variety of thresholds used in the literature, the lack of an agreed time-point for the baseline readings, and variation in the reported follow-up. With this in mind, overall reported values have ranged between 2% and 20%,7,26–29 and we previously reported that when measured from the time of treatment, and with a dual threshold multi-parameter method, progression was between 1.9% and 2.4% two years after treatment.7
The purpose of monitoring for keratoconus progression after CXL is to determine whether the disease has stabilised or whether a repeat CXL is indicated. Multiple surgeon and patient factors can influence the decision to undergo retreatment, including patient reluctance following the experience of the first procedure with little patient-perceived benefit, insufficient residual stromal thickness, and the relative lack of clinical outcome data to support the effect of repeat CXL. We found only a small number of reports on the rate of retreatment after CXL, which varied between 4 in 131 eyes (3%) to 45 in 230 eyes (20%).25,30 Interestingly, in this series, despite a minimum value for progression at five years of 6%, only a minority of eyes (0.5%) had a repeat CXL, suggesting CXL retreatment is not a good proxy for keratometric progression and that different clinical criteria apply when considering treatment with primary CXL as opposed to repeat CXL. As yet, there is also no agreement on when to perform retreatment, although the threshold differs from primary CXL, with criteria ranging from a > 2D increase in Kmax within one year, or an increase in Kmax of > 1D at two subsequent follow-up visits more than one year following the primary treatment.25,30
The strengths of our study include the large sample size available to develop the keratometry LoAs and the large cohort with extended follow-up after CXL for validation. To reduce the risk of artefacts from the acute changes in shape that can follow CXL we have also incorporated a minimum time interval after CXL to define progression. For the first time, we have then quantified the effect of using different keratometry parameters on the estimates of progression. The limitations of this study are the high proportion of eyes with incomplete data and follow-up, a risk inherent when analysing data retrospectively. In particular, we do not know if the characteristics of the population censored from the survival curves are the same as the included data. We did not analyse corneal thickness, although a change in corneal thickness is unlikely without an anterior surface change.31 Because of the high attrition rate, our results best serve as a comparative analysis of the included parameters as opposed to a statement of their relative utility to identify the true rate of progression, which is unknown. We also include older children and young adults, although there may be differences in their risk of progression. We defined our new definitions of progression as keratometry crossing an adaptive threshold after one appointment; requiring sustained progression across at least two examinations may be more accurate. Finally, an unknown proportion of the eyes in the original cohort that had primary CXL may not have progressed even if they did not have this treatment.10 These stable eyes would then also be unlikely to progress further after their primary CXL.
In conclusion, a missed diagnosis of keratoconus progression can lead to visual loss if CXL is delayed. We have confirmed that the choice of parameter and the associated threshold used to define progression will affect estimates of treatment failure following CXL. With a single or double threshold for keratometry there may be misclassification of progression in more advanced disease, and repeatable changes in keratometry in early disease may be missed. We therefore suggest that adaptive thresholds be considered as a consistent and personalised method to identify keratoconus progression. For practical implementation, we provide estimates of the LoA for each diopter of keratometry (Supplementary Table 2).