Seminal Immature Germ Cells as a Predictor of Microdissection Testicular Sperm Extraction for Development of a Combined Predictive Model

Abstract

Purpose

To investigate the relationship between immature spermatogenic cells (IGC) in semen and sperm retrieval (SR) rate of microdissection testicular sperm extraction (m-TESE) in patients with non-obstructive azoospermia (NOA).

Materials and Methods

95 patients with NOA who underwent m-TESE between 2019 and 2024 were included. These patients underwent comprehensive clinical, laboratory, and histopathological evaluations. Multivariable logistic regression analysis was used to explore the association between several variables and m-TESE success rate in patients with NOA, and a predictive model was developed. Model validity and stability were assessed through five-fold cross-validation.

Results

A positive SR was observed in 40.0% of patients undergoing m-TESE. Multivariate logistic regression analysis identified mean testicular volume (TV) (p=0.023, odds ratio [OR]=1.382, 95% confidence interval [CI]=1.046–1.826, area under the ROC curve [AUC]=0.681), IGC in semen (p<0.001, OR=13.757, 95% CI=3.663–51.671, AUC=0.785), and Johnsen score (p=0.005, OR=2.642, 95% CI=1.339–5.213, AUC=0.707) as significant predictors of positive SR, with IGC demonstrating the strongest predictive value. Four predictive models were constructed using the three factors: models A (Mean TV+Johnsen score, AUC=0.726±0.039), B (Mean TV+IGC, AUC=0.859±0.021), C (Johnsen score+IGC, AUC=0.864±0.029), and D (Mean TV, Johnsen score+IGC, AUC=0.874±0.028). Models showed significant improvements in predictive performance. The validity and stability of models B, C, and D containing IGC were better than those of model A without IGC (p<0.05).

Conclusion

The presence of IGC in the semen of patients with NOA may serve as a predictive factor for successful SR using m-TESE. Models incorporating IGC significantly enhanced stability and validity, highlighting their potential clinical utility.

INTRODUCTION

The prevalence of male infertility has shown a significant upward trend, with azoospermia accounting for 10% to 15% of cases and representing a predominant etiological factor [1,2]. The condition is clinically classified into obstructive and non-obstructive subtypes, with non-obstructive azoospermia (NOA) comprising approximately 60% of cases and representing the most severe manifestation of male reproductive dysfunction [1]. Currently, microdissection testicular sperm extraction (m-TESE) combined with assisted reproductive technologies is the most efficacious method for patients with NOA to produce biological offspring. However, m-TESE is associated with a procedural failure rate of approximately 50% and carries the risk of potential surgical complications, including transient vascular compromise and prolonged suppression of testosterone [3,4]. These clinical limitations highlight the need for reliable preoperative predictive parameters to optimize patient selection, enhance therapeutic decision-making and establish realistic patient expectations. Extensive research has investigated the clinical predictors of successful sperm retrieval (SR) in patients with NOA, including age, body mass index (BMI), testicular volume (TV), serum hormone levels, and testicular histopathology. However, many of these parameters frequently demonstrate limited specificity and sensitivity, with inconsistent findings across studies, thereby reducing their predictive value for m-TESE outcomes [5,6,7]. Although testicular histopathology remains the most reliable predictor and gold standard for assessing m-TESE success [8], its invasive nature poses surgical risks and potential tissue damage, which may impair gonadal function and negatively impact SR.

Immature germ cells (IGC) in semen, which constitute 80% to 90% of the round cells observed in ejaculates [9], represent spermatogenic precursors, including spermatogonia, spermatocytes, and spermatids. These cells can be qualitative using seminal exfoliative cytology, a non-invasive cytological assessment method. Although minimal IGC presence is typically clinically insignificant, cumulative evidence suggests that elevated IGC concentrations are inversely correlated with conventional semen parameters (sperm concentration, motility, and morphology) and fertility potential [10,11,12]. This pathophysiological association may stem from testicular injury, spermatogenic maturation arrest (MA), or functional impairment, culminating in IGC apoptosis and detachment [13,14,15]. Based on these findings, preliminary investigations have proposed seminal IGC quantification as a potential biomarker for fertility assessment and infertility classification [16,17]. Notably, the relationship between seminal IGC levels and SR rates (SRR) in TESE remains underexplored. Existing studies are largely limited by small sample sizes and reliance on outdated surgical approaches, thereby limiting their clinical applicability.

In this study, we aimed to investigate the correlation between seminal IGC levels and m-TESE outcomes in patients with NOA while identifying key clinical predictors of successful SR. Utilizing these predictors, we constructed a combined predictive model to enhance diagnostic performance and evaluated its effectiveness and stability, providing evidence-based decision support for individualized treatment strategies in NOA management.

MATERIALS AND METHODS

1. Patients

In this retrospective cohort study, we analyzed 96 treatment-naïve patients with NOA who underwent m-TESE at the Institutional Review Board of Boai Hospital of Zhongshan (IRB No. KY-2021-012-46) between January 2019 and December 2024. The inclusion criteria: according to the WHO laboratory manual for the examination and processing of human semen (6th Edition) [18], no sperm was observed in the semen; no sperm was detected in three separate semen analyses, performed each at one-week intervals; no evidence of seminal tract and seminal vesicles obstruction or absence was observed on physical, seminal plasma biochemical and scrotal ultrasound examination; and complete clinical data were available as required for the study. The exclusion criteria: presence of severe somatic or mental disorders; medication use within one month prior to enrollment; and presence of 46 XX male syndrome or complete deletion in the AZFa or AZFb region.

2. Data collection

General demographic and clinical data collected included age, height, weight, and current and past medical histories. TV, hydrocele, and varicocele were assessed using physical examination and scrotal ultrasonography. Three orthogonal dimensions of each testis were measured by ultrasonography: Length (L), maximal craniocaudal distance in the longitudinal plane; Height (H), maximum vertical distance on the longitudinal plane; width (W), maximal transverse diameters in the axial plane. TV was computed via the empirical formula of Lambert: TV (mL)=0.71×L (cm)×W (cm)×H (cm) [19]. Endocrine evaluation, peripheral blood chromosomal analysis and AZF testing were performed at the initial diagnosis of NOA following a standardized clinical protocol. Blood samples for hormone level testing were collected between 07:00–10:00 AM after an overnight fast (≥8 hours), and included measurement of follicle-stimulating hormone (FSH), luteinizing hormone (LH), testosterone, estradiol, and prolactin. Semen analysis was conducted 7 days preoperatively. Complete blood count, fasting blood glucose, hepatic/renal function, electrolyte levels, and coagulation function were assessed 24 hours preoperatively.

3. Surgical procedure

All m-TESE procedures were performed by an experienced andrologist, according to the standard methods of Schlegel [20]. The tubules were excised and placed in sterile culture dishes containing sperm buffer (Follicle Flush Buffer; Cook Medical). An experienced embryologist washed and shredded the tubules to create a homogeneous suspension. The suspension was examined under an inverted microscope at 200× magnification. SR+ was defined as the presence of at least one sperm per 100 HPF. If no sperms were found, the contralateral testis was subjected to the same procedure. The procedure was terminated upon successful SR, or if further dissection posed a risk of compromising the testicular blood supply. Testicular specimens were fixed in Bouin’s solution for subsequent histopathological evaluations.

4. Identification and assessment of IGC

A total of 1 to 2 mL of semen sample was washed three times with physiological saline (600×g/10 min). The resulting pellet was resuspended in 0.2 mL residual saline. Two smears were prepared from the resuspended pellet. Slides were air-dried and then fixed appropriately for the subsequent staining procedure. Standard Papanicolaou staining in an automated sperm stainer (DB7201, Sas Medical) to ensure batch consistency and minimize operator-dependent variability. Both smears were comprehensively examined under 1,000× oil immersion. Cell typing was performed based on cytoplasmic/nuclear staining patterns, size, and morphological characteristics [18]. The presence of IGC was documented qualitatively, includes spermatogonia, Spermatocytes and Spermatids. All slides were evaluated by two experienced cytopathologists trained in spermatogenic cell morphology.

5. Histopathological analysis

Serial 5 µm frozen sections were prepared, and processed using a standardized HE staining protocol. A single experienced pathologist evaluated the histopathological pattern of spermatogenesis, including normal spermatogenesis, hypo-spermatogenesis (HS), MA, and Sertoli cell-only (SCO) pattern. The Johnsen score was applied to each slide to assess spermatogenesis levels [2].

6. Screening predictors and constructing predictive models

Non-parametric tests were used to evaluate the differences in routine clinical parameters between the SR- and SR+ groups. Univariate (UVA) logistic regression analysis was initially performed to explore potential SR+ predictors. Multivariate (MVA) logistic regression analysis was subsequently conducted to adjust for confounding factors, ultimately identifying variables that had a significant impact on SR outcomes (p<0.05). Based on these predictors, single-factor and multi-factor predictive models were developed using logistic regression. Receiver operating characteristic (ROC) curves were generated, and the model performance was assessed using the area under the ROC curve (AUC), sensitivity, specificity, positive predictive value, and negative predictive value.

7. Internal validation and evaluation of combined predictive model

The multi-factor combined predictive model was internally validated and evaluated using five-fold cross-validation with Python software (Python 3.13.0). This validation method randomly categorized all the study participants into five equal groups, four of which were used as training sets for model building and the remaining one as a validation set for performance indicator evaluation. This process was repeated five times, with each iteration using a different group as the validation set. The final evaluation metrics were reported as the mean±standard deviation from the five validation cycles.

8. Statistical analysis

Patients were classified into SR- or SR+ groups based on m-TESE outcomes. Continuous variables were summarized as median and interquartile ranges, while categorical variables were reported as frequencies and percentages, respectively. The Mann–Whitney U test and chi-square test were used to compare continuous and categorical variables, respectively, to explore baseline data distribution differences between SR- and SR+ groups. The chi-square test and point-biserial correlation analysis were used to explore the correlation between IGC and histopathology (pathological classification and Johnson score), p-values of multiple comparisons were adjusted by Bonferroni. Logistic regression was employed to screen predictive factors and construct combined predictive models. Multiple AUC Comparisons were conducted using DeLong's non-parametric approach in the pROC package of the R (version 4.5.0) programming language, p-values were adjusted by Benjamini-Hochberg. The model underwent internal validation via five-fold cross-validation. Stability was represented by the standard deviation, while effectiveness (AUC and accuracy) was compared using the T-test. All statistical tests were two-tailed, with a significance set at 0.05. Statistical analyses were conducted using SPSS version 26.0 for Windows (IBM GmbH).

9. Ethics statement

The present study protocol was reviewed and approved by the ethics committee of Boai Hospital of Zhongshan (KY-2021-012-46). Informed consent was obtained by all subjects when they were enrolled.

RESULTS

1. Patient demographics and baseline characteristics

Table 1 summarizes the patient demographics, clinical characteristics, and laboratory parameters. Among 95 patients, 57 (60.0%) were SR- and 38 (40.0%) were SR+. No significant differences were observed between groups in age, BMI, medical history, infertility duration, semen parameters, serum sex hormones, complete blood count, blood glucose, electrolytes, hepatic/renal function, or coagulation function. Patients with SR+ exhibited significantly larger TV than SR- patients (left TV, p=0.006; right TV, p=0.003; mean TV, p=0.003). Varicoceles were present in 24 patients (the side and degree are presented in Supplement Table 1) and hydroceles in eight patients, with no significant intergroup differences. IGC were detected in the semen of 40 patients, including 11 patients with SR- (19.3%) and 29 with SR+ (76.3%) (p<0.001). In the SR- group, only spermatogonia (n=2) and spermatocytes (n=10) were observed, with no spermatids detected. By contrast, patients with SR+ exhibited spermatogonia (n=8), spermatocytes (n=29), and spermatids (n=14), with significant differences in IGC developmental stage distributions between groups (p<0.05). Histopathological analysis revealed that HS was significantly more frequent among patients with SR+ (28.9% vs. 5.3%, p=0.001). Although patients with SR- had higher rates of MA, and SCO syndrome, these differences were not significant. The Johnson scores, which reflect spermatogenic activity, were significantly higher in patients with SR+ (p=0.001).

Table 1. Demographic and clinical routine parameters characteristics of patients with non-obstructive azoospermia.

Variable		Whole cohort	Sperm retrieval negative (SR-)	Sperm retrieval positive (SR+)	p-value
Demographic and baseline characteristic
	Number of patients	95 (100)	57 (60.0)	38 (40.0)	<0.001**
	Age (y)	31 (29–35)	30 (29–34)	32 (29–35)	0.092
	Height (m)	1.71 (1.68–1.75)	1.71 (1.68–1.73)	1.72 (1.68–1.75)	0.451
	Weight (kg)	67 (60–75)	65 (60–75)	70 (63.5–77)	0.113
	BMI (kg/m2)	23.18 (21.64–25.31)	22.23 (21.36–25.25)	24.08 (22.39–25.69)	0.051
	Hypertension	16 (16.8)	10 (17.5)	6 (15.8)	0.823
	Diabetes	18 (18.9)	11 (19.3)	7 (18.4)	0.915
	History of smoking	51 (53.7)	30 (52.6)	21 (55.3)	0.801
	History of allergies	13 (13.7)	9 (15.8)	4 (10.5)	0.465
	Infertility length (y)	3 (1–5)	2 (1–4)	3.5 (2–6.25)	0.019
Scrotal anatomical factor
	Volume of left testis (mL)	4.77 (2.15–7.91)	4.02 (1.60–6.62)	6.74 (2.81–8.53)	0.006**
	Volume of right testis (mL)	5.25 (2.03–8.01)	4.24 (1.64–6.19)	6.34 (4.63–9.04)	0.003**
	Mean testicular volume (Prader)	5.25 (2.01–8.05)	4.11 (1.64–6.16)	6.99 (4.26–8.78)	0.003**
	Varicocele	24 (25.3)	14 (24.6)	10 (26.3)	0.847
	Hydrocele	8 (8.4)	6 (10.5)	2 (5.3)	0.470
Semen parameter
	Days of abstinence (day)	4 (3–5)	4 (3–5.5)	4 (3–5)	0.433
	Semen volume (mL)	3 (2.3–4)	3.1 (2.35–4)	3 (2.2–4.15)	0.828
	Liquefaction time (min)	30 (25–30)	30 (25–30)	30 (25–30)	0.510
	pH value	7.5 (7–7.5)	7.5 (7–7.5)	7.5 (7–7.5)	0.383
	Abnormal viscosity	19 (20.0)	10 (17.5)	9 (23.7)	0.464
	Leukocyte (×106/mL)	0 (0–1.5)	0 (0–1.75)	0 (0–1.75)	0.884
	Alpha-Glucosidase (U/L)	182.11 (133.64–226.65)	175.98 (119.05–220.215)	190.78 (150.345–234.615)	0.271
	Fructose (mmol/L)	23.49 (18.16–27.40)	22.1 (17.995–26.155)	24.48 (18.9775–28.31)	0.164
	Zn (mmol/L)	2.16 (1.46–2.78)	2.34 (1.53–2.785)	1.9 (1.345–2.7375)	0.376
Immature germ cells in semen
	Immature germ cells	40 (42.1)	11 (19.3)	29 (76.3)	<0.001**
	Spermatogonia	10 (10.5)	2 (3.5)	8 (21.1)	0.006**
	Spermatocytes	39 (41.1)	10 (17.5)	29 (76.3)	<0.001**
	Spermatids	14 (14.7)	0 (0)	14 (36.8)	<0.001**
Preoperative serum sex hormones
	FSH (1.4–18.1 mUI/mL)	27.56 (14.49–42.97)	29.47 (16.78–42.11)	22.965 (10.4–46.16)	0.139
	LH (1.7–8.6 mUI/mL)	9.51 (6.69–17.79)	10.09 (7.46–17.81)	8.88 (5.23–18.02)	0.466
	tT (2.48–8.36 ng/mL)	2.98 (2.10–3.93)	2.87 (2.015–3.9)	3.265 (2.175–3.97)	0.461
	E2 (<58 pg/mL)	21.05 (17.29–29.61)	20.97 (16.56–28.41)	21.54 (18.24–34.11)	0.485
	PRL (2.1–17.7 ng/mL)	9.71 (7.42–12.83)	9.86 (8–13.01)	8.69 (6.31–11.70)	0.093
	LH/FSH	0.43 (0.312–0.55)	0.41 (0.27–0.54)	0.44 (0.35–0.61)	0.065
	T/LH	0.32 (0.13–0.50)	0.29 (0.12–0.45)	0.36 (0.15–0.74)	0.074
Complete blood count
	White blood cell (×109/L)	6.4 (5.51–7.87)	6.3 (5.48–7.275)	6.5 (5.6425–8.415)	0.203
	Red blood cell (×1012/L)	5.03 (4.79–5.44)	4.95 (4.765–5.315)	5.2 (4.815–5.5325)	0.151
	Hemoglobin (g/L)	150 (142–159)	150 (141.5–156.5)	150 (142.75–161)	0.843
	Platelet (×109/L)	235 (205–269)	232 (202.5–269)	237 (209.75–268.5)	0.533
Fasting blood glucose
	FPG (mmol/L)	5.24 (4.83–5.64)	5.22 (4.845–5.5)	5.32 (4.815–5.77)	0.635
Electrolyte levels
	K+ (mmol/L)	4.11 (3.91–4.27)	4.11 (3.91–4.22)	4.11 (3.9175–4.3225)	0.747
	Na+ (mmol/L)	139.3 (138.3–140)	139.2 (138.15–140.1)	139.4 (138.4–140.1)	0.595
	Cl- (mmol/L)	105.4 (104.2–106.5)	105.3 (104.2–106.75)	105.5 (103.775–106.5)	0.752
Hepatic/renal function
	ALT (U/L)	25 (17–40)	26 (18.5–37)	23.5 (16.75–42.5)	0.973
	AST (U/L)	19 (16–25)	19 (17–24.5)	19 (15.75–28.5)	0.930
	TBIL (µmol/L)	9.7 (8.2–13.6)	9.4 (7.95–13.2)	10.45 (8.625–13.8)	0.184
	DBIL (µmol/L)	3.8 (2.7–5)	3.8 (2.45–4.95)	3.8 (2.775–5)	0.518
	BUN (mmol/L)	4.73 (4.06–5.58)	4.76 (3.99–5.61)	4.645 (4.13–5.465)	0.787
	Cr (µmol/L)	74 (66–82)	72 (63.5–82.5)	78 (66.75–82)	0.405
	UA (µmol/L)	411 (364–487)	412 (374–490)	403.5 (356–490.25)	0.715
	TP (g/L)	72.2 (69.6–75.6)	72 (69.65–75)	73.5 (69.6–76.675)	0.590
	ALB (g/L)	46.5 (44.7–48.1)	46.7 (44.7–48.25)	46.25 (44.16–47.93)	0.425
Coagulation function
	PT (S)	10.7 (10.1–11.1)	10.6 (10.1–11.05)	10.85 (10.3–11.33)	0.254
	APTT (S)	28.2 (26.8–29.6)	28.6 (27.05–29.6)	27.75 (26.68–29.65)	0.382
	Fib (g/L)	2.52 (2.2–2.92)	2.49 (2.155–2.88)	2.565 (2.225–3.0725)	0.487
	TT (S)	18.6 (18–19.2)	18.7 (18–19.2)	18.45 (18–19.1)	0.403
Histology
	HS	14 (14.7)	3 (5.3)	11 (28.9)	0.001**
	MA	51 (53.7)	35 (61.4)	16 (42.1)	0.065
	SCO	30 (31.6)	21 (36.8)	9 (23.7)	0.176
	Johnsen score	3 (2–4)	3 (2–4)	4 (3–8)	0.001**

Continuous variables were summarized as median and interquartile ranges, while categorical variables were reported as frequencies and percentages.

FSH: follicle-stimulating hormone, LH: luteinizing hormone, tT: total testosterone, E2: estradiol, PRL: prolactin, FPG: fasting blood glucose, ALT: alanine aminotransferase, AST: glutamic oxaloacetic transaminase, TBIL: total bilirubin, DBIL: direct bilirubin, BUN: urea nitrogen, Cr: creatinine, UA: urate, TP: total protein, ALB: albumin, PT: prothrombin time, APTT: activated partial thromboplastin time, Fib: fibrinogen, TT: thrombin time, HS: hypo-spermatogenesis, MA: maturation arrest, SCO: Sertoli cell-only.

^*p<0.05, ^**p<0.01.

2. Correlation between IGC and testicular histopathology

There was a correlation between IGC and testicular histopathology (p=0.011, Cramer's V=0.307, medium effect size) (Supplement Table 2). The IGC-positive rate was significantly higher in HS compared to both MA (p=0.012, Cramer’s V=0.358, medium effect size) and SCO (p=0.030, Cramer’s V=0.390, medium effect size), but no significant difference was observed between MA and SCO (p>0.05, Cramer’s V=0.014) (Supplement Table 3).The Johnson score was statistically significantly positively correlated with the presence of IGC in semen (p=0.026, r_pβ=0.229, small effect size), with IGC-positive having higher Johnson scores than IGC-negative (4.20±2.61 vs. 3.36±1.92) (Supplement Table 4).

3. Identify predictors

Table 2 presents the results of UVA and MVA logistic regression analyses. UVA identified infertility duration (p=0.023), mean TV (p=0.002), presence of IGC in the semen (p<0.001) and Johnson score (p<0.001) as potential predictors of successful SR. After adjusting for confounding factors, MVA identified mean TV (p=0.023), presence of IGC in semen (p<0.001), and Johnson score (p=0.005) as predictors significantly associated with successful SR.

Table 2. Univariate and multivariate logistic regression analysis of potential independent predictive factors for positive sperm retrieval in nonobstructive azoospermia patients undergoing m-TESE.

Variable	UVA		MVA
	OR (95% CI)	p-value	OR (95% CI)	p-value
Age	1.068 (0.972–1.173)	0.169	-	-
Height	1.022 (0.952–1.097)	0.542	-	-
Weight	1.032 (0.995–1.071)	0.090	-	-
BMI	1.149 (0.992–1.331)	0.063	-	-
Hypertension	0.881 (0.291–2.667)	0.823	-	-
Diabetes	0.944 (0.33–2.702)	0.915	-	-
History of smoking	1.112 (0.488–2.535)	0.801	-	-
History of allergies	0.627 (0.179–2.206)	0.467	-	-
Infertility length	1.207 (1.026–1.419)	0.023*	1.091 (0.893–1.323)	0.394
Mean testicular volume	1.257 (1.089–1.45)	0.002*	1.243 (1.043–1.418)	0.015*
Varicocele	1.097 (0.428–2.81)	0.847	-	-
Hydrocele	0.472 (0.09–2.474)	0.375	-	-
Days of abstinence	0.915 (0.776–1.08)	0.295	-	-
Semen volume	0.947 (0.709–1.265)	0.712	-	-
Liquefaction time	0.995 (0.961–1.031)	0.795	-	-
pH value	0.668 (0.274–1.628)	0.375	-	-
Abnormal viscosity	1.459 (0.53–4.015)	0.465	-	-
Leukocyte	0.976 (0.797–1.196)	0.815	-	-
Alpha-glucosidase	1.002 (0.998–1.007)	0.303	-	-
Fructose	1.047 (0.986–1.112)	0.136	-	-
Zn	0.861 (0.565–1.314)	0.488	-	-
Immature germ cells	13.475 (4.977–36.482)	<0.001*	13.204 (4.236–41.163)	<0.001**
FSH	0.989 (0.966–1.013)	0.382	-	-
LH	0.993 (0.943–1.046)	0.797	-	-
tT	1.215 (0.946–1.559)	0.127	-	-
E2	1.012 (0.979–1.047)	0.481	-	-
PRL	0.973 (0.899–1.054)	0.503	-	-
LH/FSH	14.585 (1.344–158.271)	0.280	-	-
tT/LH	1.16 (0.715–1.882)	0.548	-	-
Johnsen score	1.704 (1.291–2.249)	<0.001*	1.732 (1.233–2.434)	0.002**

m-TESE: microdissection testicular sperm extraction, UVA: univariable logistic regression analyses, MVA: multivariable logistic regression analyses, OR: odds ratio, CI: confidence interval, BMI: body mass index, FSH: follicle-stimulating hormone, LH: luteinizing hormone, tT: total testosterone, E2: estradiol, PRL: prolactin.

^*p<0.05, ^**p<0.01.

4. Predictor performance evaluation and model construction

Among individual predictors, IGC demonstrated the highest diagnostic accuracy (AUC=0.785), along with balanced sensitivity (76.3%) and specificity (80.7%) (Fig. 1A). The Johnsen score exhibited exceptional specificity (98.2%) but had markedly lower sensitivity (42.1%) (Fig. 1A). However, the differences in AUC between mean TV (AUC=0.681), Johnsen score (AUC=0.707) and IGC were not statistically significant (Fig. 1A) (Supplement Table 5). We developed four combined MVA predictive models incorporating key predictors: model A, mean TV+IGC; model B, mean TV+Johnsen score; model C, IGC+Johnsen score; and model D, Mean TV+IGC+Johnsen score. Model D achieved the most robust accuracy (AUC=0.876) (Fig. 1B). Similarly, there is no statistically significant difference in AUC between model B (AUC=0.855), model C (AUC=0.865), and model D, but all were higher than the single factor (Fig. 1B) (Supplement Table 5). Interestingly, model A without IGC demonstrated the lowest discriminative ability (AUC=0.728) (Fig. 1B), it was not significantly different from single factor, and significantly lower than that of models B, C, and D, and (Supplement Table 5).

Fig. 1. Receiver operating characteristic (ROC) curves of single factors and joint multi-factors in predicting the sperm retrieval of microdissection testicular sperm extraction. (A) Single factor prediction, the black dotted line is the reference , the red line is the mean TV (AUC=0.681, 95% CI: 0.564–0.789; Sen=71.1%; Spe=71.9%; PPV=62.8%; NPV=78.9%), the green line is the IGC (AUC=0.785, 95% CI: 0.687–0.884; Sen=76.3%; Spe=80.7%; PPV=72.5%; NPV=83.6%), and the blue line is the Johnsen score (AUC=0.707, 95% CI: 0.589–0.825; Sen=42.1%; Spe=98.2%; PPV=94.0%; NPV=71.8%); (B) joint multi-factors prediction, the black dotted line is the reference , the red line is the model A (AUC=0.728, 95% CI: 0.609–0.847; Sen=65.8%; Spe=86.0%; PPV=75.8%; NPV=79.0%), the green line is the model B (AUC=0.855, 95% CI: 0.779–0.937; Sen=78.9%; Spe=78.9%; PPV=65.4%; NPV=90.7%), the blue line is the model C (AUC=0.865, 95% CI: 0.785–0.945; Sen=84.2%; Spe=78.9%; PPV=72.7%; NPV=88.2%), and the orange line is the model D (AUC=0.876, 95% CI: 0.801–0.952; Sen=86.8%; Spe=78.9%; PPV=73.3%; NPV=90.0%). Model A: Mean TV+Johnsen score; Model B: Mean TV+IGC; Model C: IGC+Johnsen score; Model D: Mean TV+IGC+Johnsen score. TV: testicular volume, IGC: immature germ cells, AUC: area under the ROC curve, Sen: sensitivity, Spe: specificity, PPV: positive predictive value, NPV: negative predictive value.

5. Internal validation and performance evaluation of models

Five-fold cross-validation was performed for models A–D to assess generalizability and validate diagnostic performance. Statistical comparisons of AUC and accuracy across the five-fold validation (Fig. 2, Supplement Table 6) revealed no significant differences among models B (AUC=0.859±0.021, accuracy=84.10%±8.88%), C (AUC=0.864±0.029, accuracy=86.01%±9.25%), and D (AUC=0.874±0.028, accuracy=84.10%±10.10%), all of which included IGC data. However, these three models significantly outperformed model A (AUC=0.726±0.039, accuracy=65.52%±5.68%), which did not include IGC data. Notably, the relatively low standard deviations of AUC with models B–D (<±3%) indicate robust reproducibility.

Fig. 2. Average receiver operating characteristic (ROC) curves of the four models obtained in five-fold cross-validation. (A–D) The gray shaded area represents the range of the next standard deviation (SD) above and below the average true positive rate (TPR). (A) model A, AUC=0.726±0.039; (B) model B, AUC=0.859±0.021; (C) model C, AUC=0.864±0.029; (D) model D, AUC=0.874±0.028. The differences in AUC and accuracy of each model in the five-fold cross-validation were compared from a statistical perspective (E, F). (E) No significant difference was found in AUC among model B, C and D, but they were all higher than model A (n = 5, ^*p < 0.05); (F) No significant difference was found in the accuracy among model B, C and D, but they were all higher than model A (n = 5, ^*p < 0.05). AUC: area under the ROC curve.

DISCUSSION

In this study, we demonstrated that seminal IGC serves as a predictor of successful SR in m-TESE. Ezeh et al [21], Amer et al [22], and Koscinski et al [23] demonstrated the predictive potential of IGC for traditional TESE outcomes. Despite early evidence, research on the use of semen IGC for the evaluation of male infertility has been limited by challenges in identification and generalizability. However, advancements in seminal cytomorphology and diagnostic techniques have enabled skilled pathologists to reliably identify IGC and classify their developmental stages under high-power microscopy using standard Papanicolaou staining [18]. Under these improved conditions, our study evaluated the predictive performance of seminal IGC for m-TESE outcomes (AUC=0.785, sensitivity=76.3%, and specificity=80.7%), confirming that patients with NOA with positive seminal IGC are more likely to achieve successful SR by m-TESE. This correlation is likely associated with the degree of testicular damage and spermatogenic dysfunction. In mild-to-moderate NOA cases, developing spermatogenic cells may undergo apoptosis and be shed into the semen following testicular damage. Conversely, cases of severe NOA or SCO syndrome produce few or no early stage spermatogenic cells, making IGC detection improbable. The correlation analysis between IGC and testicular histopathology indirectly confirmed this speculation. HS had a higher IGC positivity rate than MA and SCO, which suggested that IGC may be used to indicate some testicular pathology in NOA. Notably, our study demonstrated superior sensitivity compared to the findings of Koscinski et al [24] (76.3% vs. 59.0%) while maintaining comparable specificity (80.7% vs. 83.5%). This improvement is likely owing to two key factors: (1) improved cytological techniques enabling more accurate IGC identification and (2) m-TESE’s inherently higher SR rates than conventional TESE. Which increases diagnostic sensitivity without compromising specificity, resulting in a balanced performance profile.

The Johnsen score is a quantitative histopathological grading system for assessing spermatogenic impairment in seminiferous tubules, providing an estimate of testicular sperm production potential [25]. In our cohort analysis, the Johnsen score demonstrated predictive value for SR (AUC=0.707). Its exceptional specificity (98.2%) reaffirms testicular histopathology as the gold standard for predicting m-TESE outcomes. However, its limited sensitivity (42.1%) restricts its overall predictive performance. Although the difference in AUC was not statistically significant, the diagnostic accuracy of IGC (AUC=0.785) was numerically improved compared with the Johnsen score, with balanced sensitivity (76.3%) and specificity (80.7%). This enhanced performance likely stems from methodological differences: random testicular biopsy sampling limitations may miss focal spermatogenic areas in patients with NOA, while seminal IGC analysis could reflect global testicular function. From a clinical implementation perspective, seminal IGC cytology demonstrates two advantages: enhanced predictive accuracy and a non-invasive sampling modality.

Conventional wisdom suggests that patients with NOA with larger TV and lower FSH levels exhibit a higher SRR; however, our findings revealed a limited predictive capacity for TV (AUC=0.681) and no significant predictive value for HPG hormones. These findings align with the ongoing controversy regarding their clinical utility. Some factors may have contributed to this diagnostic ambiguity. NOA is a heterogeneous condition with a complex pathophysiology, making the association between TV, HPG axis hormone levels, and SRR variable, depending on the underlying cause. For instance, patients with Klinefelter syndrome typically have smaller TVs but higher SRRs than other NOA subtypes with similar testicular sizes [26]. Additionally, TV and hormone levels may not accurately reflect the stage of spermatogenic arrest. Serum FSH concentration correlates with the number of germ cells, but not necessarily with the presence of fully developed spermatozoa [27]. Although TV and hormone levels provide a general overview of testicular function, they cannot identify isolated regions of active spermatogenesis within the seminiferous tubules [27]. The surgical approach used for SR also influences success rates. Unlike conventional TESE, m-TESE involves a meticulous examination of the entire testicular tissue, boosting the SRR two- or three-fold, even in patients with small TVs or elevated FSH levels [2]. It is also important to note the therapeutic confounders. Preoperative drug therapies (e.g., aromatase inhibitors and hCG/FSH supplementation) alter the HPG axis parameters, potentially obscuring baseline hormonal profiles. By contrast, histopathology provides a direct assessment of testicular damage, detailing the stage and type of spermatogenic disruption, but depends on pathologist expertise and biopsy sampling accuracy. Although semen IGC analysis does not directly evaluate the testicular tissue, the presence and developmental stages of IGC correlate with the severity and progression of spermatogenic failure. This suggests that IGC analysis may offer insights similar to those of histopathology, while being less susceptible to biopsy site variability.

Current predictive parameters for m-TESE outcomes remain suboptimal, as they do not demonstrate both high sensitivity and specificity. These findings underscore the need for a MVA predictive approach. Although previous models have been developed [7,28,29,30], most lack rigorous validation of stability and predictive accuracy. In our study, we developed four predictive models based on selected factors (TV, IGC, and Johnsen score). The combinatorial models demonstrated significant improvements over individual parameters, confirming that the MVA integration effectively leverages complementary predictive strengths. Notably, IGC-containing models (B, C, and D) demonstrated superior performance (AUC 0.79–0.82 vs. 0.68 for model A, standard deviation <3%), highlighting the significant contribution of IGC to the stability and predictive accuracy of the model. However, no significant differences were observed among models B, C, and D. Given the invasive nature of the Johnsen score, model B, which relies entirely on non-invasive indicators (TV and IGC), emerged as the most practical and optimal predictive model.

Although this study represents a larger-scale investigation than previous studies, its primary limitation is the relatively small sample size. First, while earlier studies reported significant associations between SRR and factors such as age and FSH, our study failed to replicate these associations, likely due to sample size constraints. Second, although various developmental stages of IGC were detected in the semen samples, the limited sample size precluded a detailed analysis of the specific developmental stages of IGC in relation to SRR. Notably, our study was limited to the qualitative assessment of seminal IGC, while quantitative analysis using flow cytometry to establish clinically actionable thresholds may be a future research direction. Moreover, previous studies have explored additional predictive factors such as anti-müllerian hormone and inhibin B [28]; however, resource constraints precluded their inclusion in our study. These considerations limit the number of variables available for constructing a combined prediction model. Future studies should prioritize expanding the sample size and incorporating a broader range of biomarkers. Implementing multicenter collaborative studies would be a practical approach for overcoming these limitations.

CONCLUSIONS

We identified TV, IGC, and Johnsen scores as predictors of m-TESE SR success. Among these, the non-invasive biomarker IGC showed a predictive performance comparable to that of histopathological evaluation, with potentially greater robustness against confounding factors. Additionally, we developed, validated, and optimized predictive models for distinguishing SR+ from SR- cases, demonstrating that IGC integration significantly improved model stability and validity.

Supplementary Materials

Supplementary materials can be found via https://doi.org/10.5534/wjmh.250156.

Supplement Table 1

Distribution of varicocele laterality and grade in non-obstructive azoospermia

Supplement Table 2

Correlation between testicular pathological types and IGC in non-obstructive azoospermia

Supplement Table 3

Pairwise comparison of the detection rate of IGC in non-obstructive azoospermia with different testicular pathological types

Supplement Table 4

Correlation between Johnson score and IGC

Supplement Table 5

Comparison of AUC of different ROC curves

Supplement Table 6

Five-fold cross-validation results of four joint diagnostic models