Laser corneal refractive surgery is based on the use of a laser to change the corneal curvature to compensate for refractive errors of the eye.1 For small incision lenticule extraction (SMILE), the lack of automated centration and cyclotorsion control in the current version of the laser system may result in undercorrections or overcorrections (or, in general, residual refractions) greater than current excimer-based refractive surgeries. Specific analyses of the age for refractive surgery outcomes also have been reported,2–4 suggesting older patients achieve more refractive change for the same attempted dioptric correction. Several studies have assessed the effects of corneal curvature on refractive surgery,5–14 suggesting overcorrections and induced positive spherical aberration in myopia, and undercorrections and induced negative spherical aberration in hyperopia.
Specific analyses of the effects of corneal curvature on refractive surgery outcomes also have been reported,2,15,16 suggesting that preoperative keratometry may play a role in the outcome of refractive surgery, with steeper corneas achieving more myopic and less hyperopic refractive change for the same attempted dioptric correction, and loss of corrected distance visual acuity (CDVA) and decreased patient satisfaction associated with steeper corneas.
Patients and Methods
Patient Population and Examinations
Consecutive patients undergoing SMILE for myopic refractive corrections between 2015 and 2019 treated at three different private practices in Zentrum für Refracktive Chirurgie, Munster, Germany (952 eyes of 566 patients); Eyereum, Seoul, Korea (1,030 eyes of 612 patients); Bergmanclinics, Utrecht, The Netherlands (582 eyes of 346 patients) were retrospectively analyzed at 3 months, 6 months, and 1 year postoeratively, respectively.
Eyes included in the study had manifest spherical equivalent refractive error ranging from −1.00 to −11.38 diopters (D) with up to 5.00 D of astigmatism. Patient charts were reviewed for the study if they had CDVA of 20/25 or better using the Early Treatment Diabetic Retinopathy Study (ETDRS) charts or Landolt rings in an electronic monitor at constant ambient illumination, stable refraction for more than 1 year prior to the study, and discontinued contact lenses for at least 2 to 4 weeks (depending on the type of contact lens) prior to the preoperative evaluation. Patients were required to have normal keratometry and topography.
The sphere and cylinder values entered into the laser were based on the manifest refraction with nomogram adjustments based on the particular experiences at each center (but the analyses were performed as deviation from the planned correction, instead of from clinical target). All eyes underwent the refractive treatment using a 6.1- to 7.4-mm lenticular diameter. Re-treatments were not considered for this retrospective study.
Drops of topical anesthetics were instilled in the upper and lower fornices. A sterile drape covering the eyelashes and face was used to isolate the surgical field. An eyelid speculum was inserted to allow maximum exposure of the globe. Proper alignment of the eye with the laser was achieved. Patients were requested to look at a pulsing green fixation light. SMILE was performed using a VisuMax 500-KHz femtosecond laser (Carl Zeiss Meditec, Jena, Germany). The lenticule was extracted and patients received topical antibiotic drops four times a day for 1 week, corticosteroid drops four times a day tapering off in 1 week, and ocular lubricants as needed.
All data were analyzed using Excel software (Microsoft Corporation, Redmond, WA).
Table 1 presents the analyzed parameters (classified as extrinsic or intrinsic) with the corresponding mean values, standard deviations, and ranges for each center.
Analyzed Parameters, Mean ± SD (Range)
Global trends were determined segregating the cohort into thirds and comparing the deviation from the plan of the lower third (from minimum to percentile 33) and the upper third (from percentile 67 to maximum) for each parameter. This way we determined whether treatments performed in the lower end of parameter A were more prone to undercorrection/overcorrection compared to treatments performed in the upper end of parameter A. Conversely, we segregated the cohort into thirds and compared the parameters of the lower third and the upper third regarding deviation from the plan. This way we determined whether treatments resulting in the lower end of undercorrection have a lower or higher value for parameter A compared to treatments resulting in the upper end of overcorrection. A parameter was considered affecting the treatment outcomes only if both complementary analyses reached statistical significance.
The effect of the different parameters on postoperative status was assessed using univariate linear and multilinear correlations. Multilinear correlations were assessed as follows:
where attempted is the planned spherical equivalent refraction, achieved is the achieved refractive spherical equivalent refraction change, mi and ni are the partial slopes of the evaluated parameters (Pi), and b is the intercept.
The first sum represents the deviation from target as a function of the input parameters (whether any/some of the input parameters induce(s) a bias [fix diopter amount] on the outcomes), whereas the second sum represents the relative deviation from target as a function of the input parameters (whether any/some of the input parameters induce(s) a gain [percentage diopter amount] on the outcomes).
For all multilinear analyses, we started with as many degrees of freedom as available for that cohort and applied stepwise regression with backward removal of the term associated with the highest P value if that was higher than the chosen cut-off of .05.
In one center, postoperative keratometries were available for all treatments. The univariate analyses were repeated as described above to determine parameters affecting the corneal flattening.
Seasonal outcomes were evaluated stratified per year season. For that, treatments were cumulated per season. Cumulated treatment refractive outcomes (per season) were compared to the global treatment refractive outcomes. The Student's t test comparing stratified values with global values was used for the statistical analysis.
To gain robustness, all analyses were run four times, once per dataset and once with the global dataset. This way, we compared which input parameters affected all three centers (and then are likely to be inherent to the procedure) and whether some affected only one dataset (ie, are likely “surgeon” dependent). From the global dataset, we obtained a robust metric on which parameters still achieve statistical significance when the cohort is three times as large.
The significance of the correlations was evaluated considering a metric distributed approximately as t with N—degrees of freedom, where N is the size of the sample considered as number of patients (and not of treated eyes).
Unpaired t tests or analysis of variance tests were used to determine statistically significant changes. A P value of less than .05 was considered statistically significant. Data from 3 months to 1 year after SMILE are reported here.
We evaluated the influence of preoperative and intraoperative parameters on postoperative clinical outcomes of SMILE in a large multicenter (multiethnic and multigeographic) population with myopia. Remarkably, astigmatism up to 5.00 D was treated in our sample. Altogether, 16% of the treatments (432 eyes) were planned with an astigmatism higher than 1.50 D.
In our study, SMILE was safe and predictable (> 99% within 1.00 D spherical equivalent refraction, 50% eyes 20/16 or better UDVA, 90% eyes 20/25 or better UDVA). The scattergrams are slightly different, but they all show that, for example, for −8.00 D actually −7.25 to −7.50 D were achieved. The differences in accuracy are explained by the different nomograms used (ie, not the distribution of the postoperative refractions, but the difference in spherical equivalent refraction to the target).
Although nomograms were already individually applied, we have performed the analyses on a “deviation from plan” approach. In such a way, the individual nomograms applied (which may actually have been evolving during the course of the 3-year treatments) shall not play a role in the measured findings.
The global trends analysis showed that age and planned sphere are the only extrinsic parameters showing a systematic effect on the outcomes. In contrast, among the analyzed intrinsic parameters, only cap thickness did not show a systematic effect. Optical zone, cap diameter, minimum lenticule thickness, and pulse energy had a systematic effect that was consistent among all three centers.
The univariate analysis confirmed these findings one-to-one and allowed us to quantify them on an individual basis (ie, without accounting for collinearities). The univariate multilinear analysis aimed to decouple effects and reduce collinearity among predictors.
The detailed values are less consistent among centers (possibly due to slight differences in technical performance among the devices), but still show that there is a global undercorrection for all centers (−0.40 D on average). After taking into account all parameters, on average among the three cohorts, −0.40 D undercorrection was the difference between what has been planned and what has been achieved as defocus correction (spherical equivalent). This average under-correction was actually not observed when considering the postoperative data alone because the centers applied their own surgeon-developed nomogram.
Cap diameter, cap thickness, and minimum lenticule thickness remain important intrinsic parameters, whereas planned sphere, age, and keratometry are the extrinsic parameters affecting the outcomes. Finally, the analysis of keratometric changes also confirms that cap diameter, cap thickness, and pulse energy remain important intrinsic parameters, whereas laterality and keratometry also affect the outcomes.
Keratometry (steeper corneas inducing undercorrections), minimum lenticule thickness (thicker reducing undercorrections), and laser energy (higher reducing undercorrections) affected all four analyses, whereas age (older reducing undercorrections), planned sphere (higher inducing undercorrections), optical zone/cap diameter (collinear; larger reducing undercorrections), and cap thickness (deeper inducing undercorrections) had an affect on refractive analyses but not on the keratometric analysis. Interestingly, laterality (second eye problem) could only be detected in the keratometric analysis (single center).
Keratometries likely affect the outcomes due to a larger deformation of the cornea onto the curved (but flatter than the cornea) patient interface (higher deformations inducing undercorrections). Minimum lenticule thickness adds tissue to be removed so it is intuitive to think that this may affect the refractive effect of the extraction. Higher laser energies produce larger cavitation bubbles, so that although the center of the bubbles may lie at the same relative locations, the envelope of the affected tissue is larger, potentially affecting the refractive effect of the extraction; planned sphere and cap thickness effects may be related to the decreasing refractive index of the cornea for deeper corneal layers.
The linear models applied (both the univariate and multilinear models) accurately represented the population findings and cannot be regarded as reporting spurious correlations by chance. The correlations were 0.49 in Germany, 0.64 in the Netherlands, and 0.78 in Korea, meaning that up to 61% of the variance could be explained by the survival model. The findings were more predictable in the refraction than in keratometry readings (providing a correlation of 0.39 [ie, only 15% of the variance being explained]). It may be that the keratometry readings are simply too noisy compared to the 0.25 D steps (or 0.125 D steps) of the refraction. An artificial intelligence system or algorithm that can be used to predict or preempt the errors could further refine our findings.
Tables 2–3 show that higher correction was associated with more undercorrection. This means that, for example, the linear regression considering only spherical equivalent refraction was of the form y = m * x + b, with m < 1 and b > 0. In other words as a fictious example, for 2.00 D of planned myopia, 1.75 D were obtained (0.25 D or −13%), but for 12.00 D of planned myopia, 10.25 D were obtained (1.75 D or −15%). On the other hand, the thicker the minimum lenticule thickness was, the less undercorrection was observed (eg, 7.00 D of myopic correction may have been undercorrected by 1.50 D [21%] when treated with a 10-µm minimum lenticule thickness, but by only 0.75 D [11%] when treated with a 20-µm minimum lenticule thickness).
The effect of patients' age on postoperative status has been assessed using univariate linear and multilinear correlations. Univariate linear analyses showed residual refraction correlated to patients' age (indicating overcorrections for older patients).
Specific analyses of the age on refractive surgery outcomes also have been reported,2–4 suggesting age may play a role in the outcome of refractive surgery, and older patients more have refractive change with the same intended dioptric correction. Those studies may indicate a relevant influence of age on refractive surgery outcomes, but subtle differences may be masked by moderate corrections and small populations. In our cohort, age influenced the achieved correction. An age-based nomogram can be quantified, in either stratified or continuous form. The regression relationship is likely valid as a system-universal compensation, but may need modification for application to other systems.
The water content of the cornea decreases with age,22,23 and it is independent from the refraction or the type of refraction. The more water the cornea contains, the lower the corneal refractive index is.24–26 Accommodation decreases with age,27–29 and it is independent from the refraction or type of refraction. However, this means that typically the patients are overminused in the refractions when they are myopic.30–35
In any refractive procedure, an ideal feature would be correction efficiency irrespective of the preoperative curvature of the cornea of the patient. Many studies have shown an increase in the correction efficiency for steeper corneas in myopia correction.
Literature reports that keratometry may influence refractive outcomes.2–11 Theoretical and clinical analyses both suggest that steeper corneas induce “hyperopic shifts” (ie, myopic overcorrections and hyperopic undercorrections).12–17
All of these parameters could also be objectively assessed by analyzing changes in keratometry readings instead of changes in refraction. Yet, postoperative keratometry readings for the complete cohort were available in only one dataset.
Of note, this nomogram is quantified based on the postoperative outcomes of the pooled data from three different VisuMax laser systems.
The diurnal fluctuation of biomechanical and morphological corneal properties is well known.36–38 However, there is no information about the moment of the day in which the surgeries were performed. Pooling together 2,564 consecutive treatments during a 3-year period, it is unlikely that in different months or seasons the moment of the day in which the surgeries were performed is significantly different.
We acknowledge the importance of environmental temperature and relative humidity during surgery, but it was not recorded in a single case record in this retrospective review. We just wanted to know whether different seasons through the year lead to different refractive outcomes. If so, one of the indicators explaining the findings could be environmental temperature and relative humidity during surgery.
To make sure that the relative overcorrections or undercorrections were not related to the different spherical equivalent planned at different seasons, we looked at the slope of the correlation, which also confirmed these differences. The treatment refractive outcome of the whole year is influenced by that of a certain season, but it is a robust analysis to compare a subset to a much larger dataset including the subset. This method is valid, although it is more prone to result in type II errors (ie, non-rejection of a false null hypothesis, also known as a false-negative finding). This is a more conservative approach because the criteria for type I errors are more stringent (ie, the significant findings are robust), whereas borderline differences tend to be rejected.
We could have compared the outcome of a certain season versus the other three seasons, but we thought that because continuities are expected, eliminating one season could artefactually increase the standard deviation of the means. Alternatively, we could have compared the outcome of a certain season versus each of the other three seasons (in a 4 × 4 matrix [or half-matrix] fashion), but we thought that it may be prone to spurious findings (type I error). Treatments performed in spring showed relative overcorrections of the spherical equivalent (P < .05), whereas treatments performed in summer showed relative undercorrections of the spherical equivalent (P < .05).
A limitation is the retrospective nature of the study. Further, the treatments were performed by three different surgeons in three different countries using three different VisuMax units, adding some extra variability to the cohort. Several confounding factors may be argued in our review (eg, we have considered both eyes of the patients).
Another limitation is that the observed undercorrection increased for treatments performed later in all three cohorts (−0.088 D or −1.46% extra undercorrection per year in the Korean cohort; −0.17 D or −3.64% extra undercorrection per year in the Dutch cohort; and −0.038 D or −0.95% extra undercorrection per year in the German cohort). We cannot exclude that this time dependency may be a consequence of changes in the surgical technique, but it seems an artefactual finding because the time dependency does not survive the stepwise backward elimination process.
This study demonstrated that both intrinsic and extrinsic treatment parameters affect postoperative outcomes in a subtle yet significant manner. Seasonal differences in refractive outcomes were observed among a large multicenter population. This may be related to environmental factors (temperature and/or humidity). The potential effect of these environmental variables on refractive outcomes warrants further evaluation.