The influence of keratometry on the accuracy of intraocular lens formulas in highly myopic eyes

Abstract

Background

To investigate the influence of keratometry on the predictive accuracy of modern IOL formulas in determining optimal IOL power for cataract patients with high myopia.

Methods

A total of 302 eyes from highly myopic patients who had undergone cataract surgery were divided into subgroups based on axial length AL and K. The predictive outcomes were to compare the accuracy of the following formulas: BU II, EVO 2.0, Haigis, Hoffer Q, Hoffer-QST, Holladay 2, K6, Kane, Pearl-DGS, SRK/T. For each formula, the ME, MAE, MedAE and the percentage of eyes with PEs within ± 0.25 D, ± 0.50 D, ± 0.75 D, and ± 1.00 D were analyzed.

Results

Among all patients, the Hoffer QST formula achieved the lowest MAE (0.41 D), closely followed by EVO 2.0 (0.42 D). In the axial length subgroups, the Hoffer QST, BU II, and Haigis formulas achieved the lowest MAE (0.35 D) in the long AL group, while K6 achieved the lowest MAE (0.46 D) in the super-long AL group. For the extreme-long AL group, Hoffer QST demonstrated the lowest MAE at 0.41 D, closely followed by EVO 2.0 (0.44 D). In the keratometry subgroups, Hoffer QST consistently showed superior accuracy: in the flat-K group, it attained the lowest MAE (0.35 D) with BU II closely trailing (0.36 D); in the average-K group, it maintained the lowest MAE (0.41 D) followed by EVO 2.0 (0.43 D); and in the steep-K group, it achieved the lowest MAE (0.43 D), closely followed by EVO 2.0 (0.44 D).

Conclusions

In highly myopic eyes, modern IOL formulas demonstrate significantly superior predictive accuracy compared to traditional formulas for the different keratometry, with Hoffer QST and EVO 2.0 exhibiting the best performance in this study.

Background

Myopia, a world spread refractive error, has emerged as a critical global public health challenge, impacting a substantial proportion of the world’s population [1, 2]. Global myopia prevalence is projected to rise from 22.9% (2000) to 49.8% (2050), marking a 117% increase over 50 years [3]. High myopia, defined as a refractive error ≤ − 6.00 diopters (D) and typically characterized by AL ≥ 26.0 mm, constitutes an independent risk factor for the development of cataract, the leading cause of vision impairment and blindness globally [3–5]. Advances in phacoemulsification and IOL designs have transformed cataract surgery from rehabilitative to refractive procedures [6]. This evolution heightens the imperative for precise postoperative refraction prediction, since even minor IOL power errors can significantly compromise visual quality and patient satisfaction [7, 8]. Consequently, the identification of optimal IOL formulas has become a central focus in both research and clinical practice.

Traditional IOL formulas generally refer to third-generation formulas, such as Hoffer Q and SRK/T [9]. These are thin-lens vergence formulas that hinge on the ELP, which is generally determined solely by AL and K [10]. Although AL and K are crucial factors in IOL power calculation, relying solely on these two parameters to calculate ELP has inherent limitations. For decades, numerous studies have shown that traditional formulas can achieve satisfactory refractive outcomes in eyes with normal AL [11]. However, for highly myopic eyes, which have more complex anatomical structures, the accuracy of traditional formulas significantly decreases when the AL exceeds 26 mm [12–14]. Therefore, a new generation of IOL formulas has emerged. These modern IOL formulas incorporate complex and diverse principles, including thick-lens formulas (e.g., EVO 2.0), ray-tracing formulas (e.g., BU II), and AI-based formulas (e.g., Hoffer QST, Kane, Pearl-DGS) [6, 15, 16].

However, few studies exist specifically on the predictive outcomes of these latest IOL formulas in highly myopic eyes under different K conditions. Therefore, we compared the performance of six modern formulas (BU II, EVO 2.0, Hoffer-QST, K6, Kane and Pearl-DGS) against four traditional formulas (Haigis, Hoffer Q, Holladay 2 and SRK/T). By analyzing the prediction errors, this study aimed to determine the influence of keratometry on modern formulas for superior predictive accuracy in highly myopic cataract patients.

Methods

Participants

This is a retrospective consecutive case series study. We reviewed consecutive patients who underwent cataract surgery between January 2024 and January 2025 in Beijing Tongren Hospital, Capital Medical University. All patients underwent phacoemulsification surgery with implantation of a one-piece hydrophobic IOL [AcrySof SN60WF IOL (Alcon, Fort Worth, TX, USA)]. Inclusion criteria were: patients who (1) completed a successful preoperative biometric examination with an AL ≥ 26 mm, (2) achieved a postoperative CDVA of ≥ 20/40. Exclusion criteria included: any history of ocular surgery, ocular pathology affecting the accuracy of biometry measurements, or perioperative complications that could compromise visual outcomes. This retrospective study complies with the Declaration of Helsinki and its subsequent revisions. Informed consent for data analysis was obtained from all patients. Ethical clearance was obtained, and all procedures in this human participant study complied with the ethical standards of the Institutional Review Board of Beijing Tongren Hospital (TRECKY2020-069), Capital Medical University.

Preoperative and postoperative assessment

All patients underwent comprehensive preoperative examinations, including UDVA, CDVA, intraocular pressure measurement, subjective refraction, as well as slit-lamp biomicroscopy. Biometric parameters were measured using the IOL Master 700 (Carl Zeiss Meditec AG), including AL, K, ACD (measured from corneal epithelium to lens), CCT, LT, and HCD, underwent subjective refraction 4 to 12 weeks postoperatively.

Surgical procedure

All patients underwent a routine phacoemulsification procedure featuring a 3.0 mm clear corneal incision, followed by intracapsular IOL implantation, all performed by experienced cataract surgeons. All surgeries for the patients were successfully completed without any intraoperative or postoperative complications.

Groupings

Based on categorization methods from previous studies [17], eyes were stratified into subgroups by AL and average K. AL subgroups were defined as: long (26.0 ≤ AL ≤ 28.0 mm), super-long (28.0 < AL ≤ 30.0 mm), and extreme-long (AL > 30.0 mm). K subgroups were established following the methodology of Röggla et al., [18] based on the mean K ± 1 standard deviation (SD) of the overall cohort: flat (K ≤ 42.58 D), moderate (42.58 D < K ≤ 45.62 D), and steep (K > 45.62 D).

IOL power calculation formulas

Preoperative ocular biometry was conducted using the IOL Master 700. The Barrett Universal II formula was employed for preoperative IOL power determination. IOL power calculations were performed using the following formulas: BU II, EVO 2.0, Haigis, Hoffer Q, Hoffer-QST, Holladay 2, K6, Kane, Pearl-DGS, SRK/T. Among these, the BU II, Haigis, Hoffer Q, Holladay 2 and SRK/T formulas were computed directly through the built-in software of the IOL Master 700. The remaining five formulas were calculated via their respective online platforms: EVO 2.0 (https://www.evoiolcalculator.com), Hoffer QST (https://www.hofferqst.com), K6 (https://www.cookeformula.com), Kane (https://www.iolformula.com) and Pearl-DGS (https://www.iolsolver.com). The optimized lens constants were obtained from the IOLCon website (https://www.iolcon.org).

Outcome measures

In accordance with the latest IOL formula study protocol, PE serves as the metric for evaluating refractive outcome accuracy. It is defined as the difference between the actual postoperative SE and the predicted SE based on a specific formula. A negative PE represents a myopic shift, whereas a positive PE represents a hyperopic shift. For each formula, the following metrics were calculated: ME, MAE, MedAE, and the proportion of eyes achieving PEs within ± 0.25 D, ± 0.50 D, ± 0.75 D, and ± 1.00 D [19, 20].

Statistical analysis

Statistical analysis was performed using SPSS 26.0. The normality of continuous variables was assessed using the Shapiro-Wilk test. Differences in PE and AE among the five IOL formulas were compared using the Friedman test. Subsequently, pairwise comparisons were conducted using the Wilcoxon signed-rank test, with outcomes adjusted through the Bonferroni correction (α = 0.05). Categorical variables were presented as frequencies (%). The Cochran Q test was used to compare the proportions of eyes achieving accuracy thresholds (≤ 0.25 D, ≤ 0.50 D, ≤ 0.75 D, and ≤ 1.00 D) across the IOL formulas. P-values less than 0.05 were regarded as statistically significant.

Results

This study included 302 eyes from 302 consecutive patients with a mean age of 57.7 ± 9.8 years. The mean AL was 29.28 ± 2.26 mm, mean K was 44.18 ± 1.49 D, and mean ACD was 3.47 ± 0.30 mm. The implanted IOL power ranged from − 5.0 to + 16.0 D, with an average power of 5.91 ± 4.65 D. Demographic and ocular biometric characteristics of the cohort were detailed in Table 1.

Table 1.

The demographics and ocular biometric data of the study cohort

Parameter	Mean ± SD/ Median (Q1, Q3)	Minimum	Maximum
Gender (Male/Female)	104/198	-	-
Age(Y)	57.7 ± 9.8	29	83
Postoperative SE (D)	-0.75 (-1.375, -0.5)	-6.625	0
Biometry
Axial length (mm)	29.28 ± 2.26	26.00	36.71
Mean keratometry (D)	44.18 ± 1.49	39.62	49.85
Flat keratometry (D)	43.53 ± 1.50	39.20	48.77
Steep keratometry (D)	44.68 ± 1.63	40.04	50.83
Anterior chamber depth (mm)	3.47 ± 0.30	2.67	4.42
Lens thickness (mm)	4.48 ± 0.38	3.40	5.98
Corneal central thickness (µm)	518.7 ± 34.8	423	603
Horizontal corneal diameter (mm)	11.7 (11.3, 12.0)	10.6	12.9
IOL power (D)	6 (2, 10)	-5	16

Anterior chamber depth, as measured from corneal epithelium to lens anterior surface

SD = Standard deviation; Q1 = 25% percentile; Q3 = 75% percentile; SE = spherical equivalent; D = diopter; IOL = intraocular lens

The postoperative refractive outcomes of 10 formulas were summarized. BU Ⅱ, K6, Kane, and Pearl-DGS resulted in slight myopic shifts (MEs ranging from − 0.16 to -0.08 D), whereas EVO 2.0, Haigis, Hoffer Q, Hoffer QST, Holladay 2, and SRK/T formulas produced hyperopic shifts (MEs ranging from 0.01 to 0.72 D). Overall, significant differences in predictive accuracy were observed among the formulas. Modern formulas demonstrated significantly lower MAE and higher proportions of eyes with PEs within ± 1.00 D compared to traditional formulas (P < 0.01). Among them, Hoffer QST (0.41 D) and EVO 2.0 (0.42 D) demonstrated the lowest MAE, with no significant difference between them (P = 0.10). However, despite both of them leading to minor hyperopic shifts, there was a statistically significant difference in their MEs (P = 0.01) (Table 2).

Table 2.

Postoperative refractive outcomes of all patients

Formula	ME ± SD	MAE	MedAE	Percentage of Eyes Within Different PE(%)
				± 0.25	± 0.50	± 0.75	± 1.00
Barrett UⅡ	-0.09 ± 0.56	0.44	0.35	34.77	66.89	81.46	92.72
EVO 2.0	0.01 ± 0.54	0.42	0.36	38.08	67.22	84.44	94.04
Haigis	0.25 ± 0.54	0.47	0.38	36.42	61.59	81.13	91.06
Hoffer Q	0.72 ± 0.66	0.8	0.72	18.22	37.42	51.32	68.21
Hoffer QST	0.03 ± 0.53	0.41	0.33	40.07	69.21	85.1	93.71
Holladay 2	0.55 ± 0.62	0.67	0.6	21.52	41.06	64.24	78.81
Cooke K6	-0.08 ± 0.56	0.44	0.37	34.11	66.56	82.45	93.38
Kane	-0.16 ± 0.60	0.47	0.38	34.77	64.24	80.46	90.4
Pearl-DGS	-0.08 ± 0.58	0.45	0.36	35.76	63.91	83.44	91.72
SRK/T	0.23 ± 0.70	0.56	0.46	27.81	53.31	72.85	87.42

SD = standard deviation; PE = prediction error; ME = mean prediction error; MAE = mean absolute error; MedAE = median absolute error

The postoperative refractive outcomes varied among different AL groups (Table 3). The ME and MAE of all formulas changed with increasing AL in Fig. 1A and B. The modern formulas exhibited significantly greater stability than traditional formulas as AL varies. The P-values of all modern formulas were > 0.05, except for Kane (P < 0.01). In contrast, the P-values of all traditional formulas were < 0.01(Fig.  2A-J). The proportion of eyes with PEs within ± 1.00 D of 10 formulas across AL subgroups was shown in Fig. 3A. The proportion within ± 1.00 D was consistently higher for modern formulas compared to traditional formulas in each AL subgroup.

Table 3.

Postoperative refractive outcomes in different AL subgroups

Formulas	26 ≤ AL ≤ 28 mm (n = 113)		28< AL ≤ 30 mm (n = 86)		AL>30 mm (n = 103)
	ME ± SD	MAE	ME ± SD	MAE	ME ± SD	MAE
Barrett UⅡ	-0.08 ± 0.44	0.35	-0.22 ± 0.63	0.53	0.02 ± 0.59	0.46
EVO 2.0	-0.01 ± 0.47	0.38	-0.03 ± 0.59	0.46	0.04 ± 0.58	0.44
Haigis	0.07 ± 0.42	0.35	0.17 ± 0.58	0.47	0.50 ± 0.52	0.59
Hoffer Q	0.31 ± 0.45	0.45	0.64 ± 0.63	0.74	1.24 ± 0.54	1.24
Hoffer QST	0.05 ± 0.44	0.35	-0.11 ± 0.6	0.47	0.13 ± 0.53	0.41
Holladay 2	0.23 ± 0.44	0.42	0.49 ± 0.62	0.64	0.99 ± 0.52	1.05
Cooke K6	-0.08 ± 0.50	0.4	-0.06 ± 0.59	0.46	-0.09 ± 0.59	0.48
Kane	-0.02 ± 0.48	0.39	-0.13 ± 0.6	0.48	-0.34 ± 0.66	0.55
Pearl-DGS	-0.02 ± 0.48	0.39	-0.14 ± 0.62	0.49	-0.09 ± 0.63	0.43
SRK/T	-0.12 ± 0.50	0.4	0.15 ± 0.61	0.49	0.68 ± 0.71	0.79

AL = axial length; ME = mean prediction error; MAE = mean absolute error; SD = standard deviation

Fig. 1.

Smoothed curve analysis of ME and MAE for ten IOL formulas. D = diopters. D = diopters

Fig. 2.

Correlation of AL and K with prediction errors in ten IOL formulas. D = diopters

Fig. 3.

Percentages of eyes within ± 0.25 D, ± 0.50 D, ± 0.75 D, and ± 1.00 D prediction error for ten IOL formulas across subgroups. D = diopters

The performance of refractive outcomes in different K groups was presented in Table 4. The ME and MAE of all formulas changed with increasing K in Fig. 1C and D. The variations in K had minimal impact on the predictive accuracy of IOL formulas. The P-values were > 0.05 for all formulas except K6 (P < 0.01) (Figs. 2K-T). The proportion of eyes with PEs within ± 1.00 D of 10 formulas across K subgroups was shown in Fig. 3B. Within each K subgroup, the proportion within ± 1.00 D was consistently higher for modern formulas compared to traditional formulas.

Table 4.

Postoperative refractive outcomes in different keratometry subgroups

Formulas	K ≤ 42.58 D (n = 52)		42.58 D< K ≤ 45.62 D (n = 205)		K>45.62 D (n = 45)
	ME ± SD	MAE	ME ± SD	MAE	ME ± SD	MAE
Barrett UⅡ	0.07 ± 0.49	0.36	-0.10 ± 0.55	0.44	-0.21 ± 0.61	0.53
EVO 2.0	0.13 ± 0.51	0.39	-0.02 ± 0.54	0.43	-0.04 ± 0.55	0.44
Haigis	0.31 ± 0.50	0.43	0.23 ± 0.54	0.48	0.27 ± 0.56	0.46
Hoffer Q	0.72 ± 0.66	0.78	0.71 ± 0.67	0.8	0.78 ± 0.66	0.83
Hoffer QST	0.05 ± 0.48	0.35	0.02 ± 0.53	0.41	0.06 ± 0.56	0.43
Holladay 2	0.71 ± 0.49	0.82	0.52 ± 0.64	0.67	0.58 ± 0.63	0.66
Cooke K6	0.22 ± 0.53	0.44	-0.13 ± 0.55	0.44	-0.21 ± 0.53	0.46
Kane	0.07 ± 0.49	0.38	-0.21 ± 0.61	0.48	-0.20 ± 0.61	0.52
Pearl-DGS	0.02 ± 0.50	0.37	-0.10 ± 0.58	0.46	-0.12 ± 0.62	0.50
SRK/T	0.47 ± 0.54	0.57	0.19 ± 0.71	0.56	0.12 ± 0.73	0.52

K = keratometry; ME = mean prediction error; MAE = mean absolute error; SD = standard deviation

The lowest MedAE of each formula in the defined subgroups was shown in Table 5. In eyes with long AL and flat K, the Hoffer QST formula yielded the lowest MedAE (0.13 D), closely followed by Haigis (0.15 D). For eyes with long AL and average K, the Haigis formula achieved the lowest MedAE (0.27 D), closely followed by Hoffer QST (0.29 D). In eyes with long AL and steep K, BU II formula demonstrated the best performance (0.26 D). For eyes with super-long AL and either flat or average K, the Kane formula demonstrated superior accuracy, whereas eyes with super-long AL and steep K were better accommodated by the Pearl-DGS formula. Notably, for the eyes with extreme-long AL, the Hoffer QST formula was the best advisable selection.

Table 5.

Median absolute error of each formula

Lowest MedAE	Long AL	Super-long AL	Extreme-long AL
Flat K	n = 13 1. HQST (0.13) 2. Haigis (0.15) 3. DGS (0.19)	n = 20 1. Haigis, Kane (0.22) 2. K6, EVO (0.23) 3. HQST (0.27)	n = 19 1. BU Ⅱ (0.31) 2. HQST (0.33) 3. Kane (0.36)
Average K	n = 79 1. Haigis (0.27) 2. HQST (0.29) 3. K6 (0.33)	n = 50 1. Kane (0.41) 2. K6, HQST (0.43) 3. DGS (0.44)	n = 76 1. DGS (0.31) 2. HQST (0.32) 3. EVO (0.33)
Steep K	n = 21 1. BU Ⅱ (0.26) 2. K6, SRK/T (0.27) 3. DGS (0.33)	n = 16 1. DGS (0.32) 2. Haigis (0.33) 3. HQST, K6 (0.34)	n = 8 1. Haigis (0.26) 2. HQST (0.33) 3. SRK/T (0.72)

AL = axial length; K = keratometry; MAE = mean absolute error; BU Ⅱ=Barrett UⅡ; K6 = Cooke K6; HQST = Hoffer QST; DGS = Pearl-DGS; EVO = EVO 2.0

Discussion

This study analyzed the accuracy of IOL formulas for cataract patients with high myopia, and observed the postoperative refractive prediction under different keratometry. The research demonstrated that all IOL formulas could accurately predict the postoperative refractive status. However, in the extreme-long AL group, the modern formulas exhibited better predictive accuracy compared to traditional formulas. In different keratometry groups, steep K had a relatively minor impact on the predictive accuracy of modern formulas. For the extreme-long AL the Hoffer QST formula demonstrated good stability under different keratometry. These findings had significant clinical relevance for surgeons intending to implant IOLs in high-myopia patients.

Highly myopic eyes exhibited complex anatomical changes, such as increased AL, deeper ACD, posterior staphyloma, lax zonules, and reduced capsular bag support [21, 22]. These factors collectively impacted the reliability of IOL formulas. As a representative fourth-generation formula, the Barrett Universal II was based on ray-tracing methods, which calculated the path of individual light rays through optical interfaces using Snell’s law and the refractive induced of each medium. Numerous studies had confirmed its superior predictive accuracy in highly myopic eyes compared to traditional formulas [23–25]. In recent years, several new modern formulas based on different principles had emerged, including thick-lens vergence formulas (EVO 2.0 and Pearl-DGS) and formulas incorporating AI algorithms (Hoffer QST and Kane). Recent research had indicated that the performance of these new-generation formulas formulas was comparable or even better compared to the BUII [26]. Darcy et al. [27] demonstrated the Kane formula’s superior accuracy across all axial length subgroups and the overall cohort. Concurrently, Diogo et al. [28] confirmed the reliability and stability of the new-generation formulas, particularly Kane, Pearl-DGS, and EVO 2.0.

Generally, the accuracy of modern formulas was superior to that of traditional formulas. In this study, modern formulas consistently demonstrated satisfactory predictive accuracy, predominantly showing a tendency towards myopic shift. With the exception of Kane (0.16 D), the MAE of the other five modern formulas was within 0.50D. Among them, Hoffer QST (0.41 D) and EVO 2.0 (0.42 D) demonstrated the best performance. All modern formulas achieved a rate of ≥ 90% of the proportion of eyes with PEs within ± 1.00 D. Hoffer QST (93.71%) and EVO 2.0 (94.04%) were the best formulas. These findings were consistent with previous studies [28, 29]. In contrast, the predictive accuracy of traditional formulas was inferior to that of modern formulas, consistently showing a hyperopic shift. This aligns with the previously mentioned limitations inherent in the design principles of traditional formulas. Among the four traditional formulas, only Haigis (0.47 D) achieved an MAE within 0.50 D, while SRK/T (0.56 D), Holladay 2 (0.67 D), and Hoffer Q (0.80 D) showed a significant gap compared to modern formulas. Similarly, only Haigis (91.06%) achieved a rate of ≥ 90% of the proportion of eyes with PEs within ± 1.00 D. The rate for Hoffer Q was substantially lower at just 68.21%. Therefore, with the exception of Haigis, traditional formulas were not suitable for IOL power calculation in eyes with high myopia, particularly Hoffer Q [16].

Considering that AL and K are the two most significant factors influencing refractive outcomes, we conducted a further analysis of the results by categorizing eyes according to their AL and K values. Within long AL group, Barrett U Ⅱ, Hoffer QST, and Haigis achieved the lowest MAE (0.35 D). In super-long AL group, EVO 2.0 and K6 demonstrated the best performance with an MAE of 0.46 D. For extreme-long AL group, Hoffer QST again showed the lowest MAE (0.35). In the AL groups, as AL increased, the MAE of all formulas gradually increased and the predictive outcomes of the formulas exhibited a shift from myopic bias towards hyperopic bias [14, 30]. This confirms that AL was a crucial factor determining formula performance. In the K groups, variations of K did not lead to significant changes in the MAE of any formula, indicating that K had a limited impact on formula performance [25, 31]. However, when AL and K were analyzed together, the best-performing formula in defined subgroup changed markedly. In the flat K group, Hoffer QST and Haigis showed relatively good performance across the various AL subgroups. Within the average K group, Hoffer QST and Pearl-DGS exhibited relatively lower prediction errors across AL subgroups. In the steep K group, Pearl-DGS and Haigis performed comparatively well. This highlights that both the K and AL must be considered for the IOL calculation, whereas relying on only one parameter can result in a poor formula choice [18].

This study was subject to several limitations. First, due to the retrospective nature of the study, although a sufficient sample size was obtained for heteroscedastic test, the statistical power of some subgroups analyses may still be inadequate. Therefore, these results should be considered for informational purposes only. Furthermore, our research focused on a single-center Chinese population. Therefore, it was essential to validate these findings through multicenter and multiethnic studies in subsequent research.

In conclusion, the accuracy of modern formulas was superior to that of traditional formulas for the different keratometry in highly myopic eyes. Among traditional formulas, Haigis is the only recommended formula. Modern formulas generally performed well, with Hoffer QST and EVO 2.0 being the most recommended. These findings may serve as a valuable guide for cataract surgeons when selecting optimal formulas for highly myopic eyes, potentially enhancing patients’ postoperative visual satisfaction.

Abbreviations

IOL

Intraocular lens

AL

Axial length

K

Keratometry

BU II

Barrett Universal II

EVO

Emmetropia Verifying Optical

K6

Cooke K6

ME

Mean prediction error

MAE

Mean absolute prediction error

MedAE

Median absolute prediction error

ELP

Effective lens position

CDVA

Corrected distance visual acuity

UDVA

Uncorrected distance visual acuity

ACD

Anterior chamber depth

CCT

Central corneal thickness

LT

Lens thickness

HCD

Horizontal corneal diameter

PE

Prediction error

SE

Spherical equivalent

Author contributions

G.Y.L, P.Z, X.H.W and J.S.Z conceived the project, designed experiments, and wrote the manuscript. Z.Y.L, Z.H.Y, M.L and JX. collected data, analyzed data and prepared figures. All authors read and approved the final manuscript.

Funding

Fund of the Excellent Clinical Research Program for Research Wards (No. BRWEP2024W172050109).

Data availability

All data generated or analyzed during this study are included in this published article. Further enquiries can be directed to the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This retrospective study complies with the Declaration of Helsinki and its subsequent revisions. Informed consent for data analysis was obtained from all patients. Ethical clearance was obtained, and all procedures in this human participant study complied with the ethical standards of the Institutional Review Board of Beijing Tongren Hospital (TRECKY2020-069), Capital Medical University. Clinical trial number: not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.