Introduction

Autosomal dominant polycystic kidney disease (ADPKD) is the fourth leading cause of end-stage kidney disease (ESKD), accounting for 5–10% of cases1. Caused by mutations in the PKD1 or PKD2 genes, which encode polycystin-1 and − 2, ADPKD involves the gradual development of large, fluid-filled cysts in the kidneys over decades2, often leading to various complications3,4,5,6. Cysts in both kidneys increase in size and number with age, causing progressive renal function decline, with approximately half of ADPKD patients reaching end-stage kidney disease (ESKD) by their sixties7. The renal function typically starts declining in the fourth decade, with an estimated glomerular filtration rate (eGFR) decrease of 4.4 to 5.9 mL/min per year from that point8.

Many risk factors for renal function decline have been reported in patients with ADPKD9,10,11,12,13, including PKD1 gene mutations, truncating mutations, splicing/frameshift mutations, male sex, hypertension, hyperphosphatemia, proteinuria, and total kidney volume (TKV).

Although anemia commonly appears in chronic kidney disease (CKD) and is associated with poor renal prognosis14,15,16, it occurs less frequently in ADPKD due to higher hemoglobin levels maintained by erythropoietin (EPO) production from cysts17,18,19. Consequently, the impact of anemia on renal prognosis in ADPKD has not been extensively studied, with only a few small studies investigating its association with disease progression11,20,21. Therefore, we analyzed anemia as a prognostic factor in a larger cohort of Japanese patients with ADPKD.

Methods

Study design

We reviewed the records of 586 outpatients with ADPKD who visited the Kidney Center at Tokyo Women’s Medical University Hospital (Tokyo, Japan) between July 2003 and July 2019. Patients who were already undergoing dialysis therapy (n = 33) were excluded, resulting in a final cohort of 553 patients (Supplementary Fig. 1). ADPKD was diagnosed using previously described criteria22. Clinical data were collected from baseline examinations, and the patients were prospectively observed to evaluate outcomes. All procedures were approved by the Research Ethics Committee of Tokyo Women’s Medical University (No. 5118) according to the 1964 Declaration of Helsinki and its later amendments or equivalent ethical standards. Passive informed consent (opt-out) was obtained from patients, and all data were analyzed anonymously. Additional methodological details are provided in the supplementary material (Supplementary Methods). Baseline antihypertensive medication use was recorded and is presented in Table 1 with subclassification into renin–angiotensin system inhibitors (ACE inhibitors and/or angiotensin receptor blockers) and calcium channel blockers, overall and by sex. Anemia was classified into five categories based on hemoglobin (Hb) levels: <10 g/dL, < 11 g/dL, < 12 g/dL, < 13 g/dL, and < 14 g/dL, with anemia defined by Hb levels in these ranges or by the use of iron or erythropoiesis-stimulating agents (ESAs). Patients using tolvaptan were excluded from the renal prognosis analysis, and those who initiated tolvaptan during follow-up were censored. The participants were followed up until October 31, 2020.

Table 1 Patient characteristics (entire cohort, and sub-cohorts based on sex).
Full size table

Outcome evaluations

Renal outcome was defined as a 50% reduction in eGFR or the initiation of renal replacement therapy.

Statistical analyses

Continuous variables were expressed as means ± standard deviations or medians (range), whereas discrete variables were expressed as percentages. For continuous variables, the Mann–Whitney U test or unpaired t-test was performed as appropriate after assessing data normality. The chi-square or Fisher’s exact test was used to analyze categorical variables.

Logistic regression models were fitted at each follow-up year to evaluate the association between anemia definitions and the renal outcome. Model fit was summarized using McFadden’s pseudo-R², defined as pseudo-R² = 1 − (lnL_model / lnL_null), where lnL_model is the log-likelihood of the fitted model and lnL_null is that of the intercept-only model23. Pseudo-R ² is a relative goodness-of-fit index (improvement over the null model) and is not interpreted as an absolute proportion of variance explained. In this study, we used pseudo-R ² primarily to compare the relative prognostic contribution of different Hb cut-offs over time (“time-series pseudo-R²” 24–26), rather than to claim high absolute predictive performance. We further summarized the time-series curves by calculating the mean and the integral of pseudo-R² over prespecified periods (e.g., years 1–17, years 1–END) to rank Hb cut-offs for long-term prognosis (1–17Y integrals/1–17Y means, 1Y–END integrals/1Y–END means). Multivariable Cox regression was additionally performed to confirm associations using a conventional time-to-event framework.

For validation, multivariable Cox regression analyses were also performed using a conventional time-to-event framework to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for the primary endpoint. Variables of interest and general risk factors for outcomes based on existing knowledge were included in the multivariable model24,25,26,27. Standard methods were used to estimate the sample size for multivariable Cox regression analyses with ≥ 5 outcomes required for each independent variable28,29. Given multiple comparisons across cut-offs and time points, we focused on consistency of patterns across the pseudo-R² summaries and Cox models, and we tempered causal or “determinant” language accordingly. Statistical significance was set at 5% to identify areas warranting further investigation. Cox regression constituted the primary inferential analysis, whereas time-series pseudo-R² was used only as a complementary, exploratory visualization tool. All tests were two-tailed, and statistical analyses were performed using JMP Pro software (version 17.0.0; SAS Institute, Cary, NC, USA).

Results

Patient characteristics

Patient characteristics are summarized in Table 1. The median ages were 43 and 45 years for women and men, respectively. Women exhibited higher median eGFR (62.0 vs. 48.1 mL/min/1.73 m², P < 0.0001), lower mean height-adjusted total kidney volume (htTKV) (650.5 vs. 996.4 mL, P < 0.0001), and lower mean Hb levels (12.0 vs. 13.6 g/dL, P < 0.0001) than men. Anemia with Hb < 13.0 g/dL was found in 51.2% of all patients, 31.6% of men, and 66.7% of women. Moreover, anemia with Hb < 12.0 g/dL was observed in 32.6% of all patients, 22.5% of men, and 40.4% of women. Iron and ESAs were used by 8.1% of all patients, including 7.0% of men and 9.1% of women. With respect to concomitant medications, antihypertensive agents were prescribed in 44.1% of the overall cohort, with a higher prevalence in men than in women (59.4% vs. 32.0%, P < 0.0001). Renin–angiotensin system inhibitors (ACE inhibitors and/or angiotensin receptor blockers) were used by 36.1% of patients overall, again more frequently in men than in women (51.2% vs. 24.3%, P < 0.0001), while calcium channel blockers were prescribed in 25.1% of patients (32.4% in men vs. 19.1% in women, P = 0.0003). These data are presented in detail in Table 1 to provide clinical context for interpretation of anemia and renal prognosis.

Renal prognostic indicators in patients with ADPKD

The median follow-up period was 9.1 years, with 157 patients reaching the primary endpoint. We analyzed renal outcomes using time-series pseudo-R2 values for each anemia definition30,31(Table 2; Fig. 1). The highest 1-17Y/1-END integral and mean pseudo-R2 values were observed for anemia defined by Hb < 12 g/dL in the entire cohort, Hb < 13 g/dL in men, and Hb < 12 g/dL in women. Consequently, Hb levels of 12 g/dL overall, with sex-specific thresholds of 13 g/dL in men and 12 g/dL in women, were identified as candidate cut-offs for anemia associated with ADPKD prognosis within the prespecified analytic framework. In particular, anemia with Hb < 13 g/dL in men showed the highest 1–17Y mean pseudo-R² value (0.1838), indicating a more persistent association with long-term renal outcomes compared with other prespecified Hb cut-offs23.

Table 2 Time-series change in pseudo-R2 values of the prognostic efficacy for renal outcomes regarding anemia (entire cohort, and sub-cohorts based on sex).
Full size table
Fig. 1
Fig. 1
Full size image

Time-series changes in pseudo-R2 values for the kidney outcomes based on different anemia definitions. (A–C) Pseudo-R2 values for all anemia definitions in the entire sub-cohort and by sex. The time-series changes in pseudo-R2 values for the entire cohort (A), men (B), and women (C) are shown with superimposed lines representing each anemia definition. (D–F) Pseudo-R2 values in terms of Hb levels. (G–I) Pseudo-R2 values for anemia defined as Hb < 10.0 g/dL. (J–L) Pseudo-R2 values for anemia defined as Hb < 11.0 g/dL. (M–O) Pseudo-R2 values for anemia defined as Hb < 12.0 g/dL. (P–R) Pseudo-R2 values for anemia defined as Hb < 13.0 g/dL. (S–U) Pseudo-R2 values for anemia defined as Hb < 14.0 g/dL. Gray areas beneath each line represent the integrated pseudo-R2 values from the first year to the last follow-up year (1Y-END integrals). Hb, hemoglobin; ESAs, erythropoiesis-stimulating agents.

Multivariable Cox analyses validated Hb (HR 0.81; 95% CI 0.72–0.92; P = 0.0016) as an independent predictor of a ≥ 50% decline in eGFR or need for renal replacement therapy (Table 3). Cox regression analysis of Hb cut-offs associated anemia with Hb < 11 g/dL with a higher risk for poor ADPKD prognosis, with the highest HR of 1.66; however, anemia with Hb < 12 g/dL yielded a similar HR of 1.62. For men, anemia defined by Hb < 13 g/dL had the highest HR of 3.36, whereas for women, Hb < 12 g/dL had the highest HR of 2.14. Therefore, multivariable analysis using anemia cut-offs of Hb < 13.0 g/dL for men and Hb < 12.0 g/dL for women showed higher HRs compared to using a universal Hb < 12.0 g/dL cut-off (HR = 2.11 vs. HR = 1.89, respectively) (Table 4). In a sensitivity analysis excluding patients treated with tolvaptan at baseline, the overall pattern of associations and the relative ranking of sex-specific Hb cut-offs were broadly consistent with the primary analysis (Supplementary Table 1).

Table 3 Multivariable cox analyses for risk factors associated with the kidney outcome according to several Hb cut-offs (entire cohort, and sub-cohorts based on sex).
Full size table
Table 4 Multivariable Cox analyses for risk factors associated with the kidney outcome according to sex-specific Hb cut-offs (entire cohort).
Full size table

Discussion

This study examined integral time-series pseudo-R2 values for Hb using various cut-offs to identify candidate anemia definitions associated with long-term kidney prognosis in patients with ADPKD. The findings identified Hb < 12 g/dL as a general threshold, with sex-specific cut-offs of Hb < 13 g/dL in men and Hb < 12 g/dL in women, which were consistently supported by multivariable Cox analyses.

The prognostic strength of kidney risk factors can depend on follow-up duration32, as demonstrated by time-series analyses of pseudo-R² values that capture temporal changes in relative model contribution. Factors exhibiting high or increasing pseudo-R2 values indicate long-term importance33, whereas those with declines may suggest short-term relevance or treatment effects30,31,33. Integral and mean pseudo-R² summaries therefore provide an intuitive framework to compare the relative contribution of prespecified risk factors under identical model specifications across follow-up periods.

Importantly, pseudo-R² was applied in this study as a relative goodness-of-fit index, rather than as a measure of absolute predictive performance or clinical utility. This approach allows exploration of how the contribution of hemoglobin varies over the disease course in a chronic and heterogeneous condition such as ADPKD, where risk factor relevance is unlikely to remain constant over time.

Anemia is common in CKD34,35 and is associated with accelerated kidney function decline and progression to ESKD14,16. However, its impact on ADPKD prognosis has been less well characterized. Prior studies have suggested associations between low hemoglobin levels and ADPKD progression, but these investigations were limited by relatively small sample sizes20,21. The present study, based on a larger cohort, extends these observations by demonstrating consistent associations between lower hemoglobin levels and adverse renal outcomes using both time-series pseudo-R² analyses and conventional survival models.

A potential biological explanation involves hypoxia-inducible factors (HIFs). In early ADPKD, cyst expansion and tissue compression induce hypoxia, activating HIF-1α and HIF-2α and increasing erythropoietin production and Hb levels36,37. As disease progresses, however, persistent HIF-1α activation promotes cyst growth38,39,40, while interstitial injury and ischemia impair erythropoietin production, leading to anemia and potentially signaling disease progression. Renin–angiotensin system inhibition has been reported to suppress erythropoietin production and erythroid progenitor activity, potentially leading to lower hemoglobin levels, while simultaneously exerting renoprotective effects41. In our cohort, the distribution of anemia did not parallel ACE-I/ARB use, supporting interpretation of anemia as a prognostic marker rather than a treatment-related effect. Although medication use was not included as a time-varying covariate, the discordant sex distribution between ACE-I/ARB use and anemia prevalence suggests that these agents are unlikely to fully account for the observed prognostic associations.

The relevance of attribute-based medicine, particularly with respect to sex differences in anemia, warrants consideration. Attribute-based medicine emphasizes precision by accounting for patient- and disease-specific attributes, among which sex differences are particularly important42,43,44. The observed sex-specific hemoglobin thresholds are consistent with this framework. From an attribute-based medicine perspective, the present findings underscore the importance of interpreting anemia not as a uniform condition, but as a prognostic signal whose clinical significance varies according to patient attributes such as sex and disease context. Attribute-based medicine aims to refine risk stratification by integrating biologically and clinically meaningful attributes, rather than applying a single threshold across heterogeneous populations. In this framework, the observed sex-specific hemoglobin cut-offs should be understood as tools for precision-oriented risk interpretation and disease understanding rather than as universal therapeutic targets. This concept is consistent with prior attribute-based studies demonstrating that constellations of patient attributes—rather than isolated risk factors—shape renal and cardiorenal prognosis across diverse kidney diseases43,45,46. Furthermore, emerging evidence suggests that age- and sex-related biological transitions, such as postmenopausal hormonal changes, may modify disease vulnerability and risk patterns in ADPKD, reinforcing the relevance of an attribute-based approach47.

Because men generally have higher baseline hemoglobin levels, a comparable absolute decline may represent a greater physiological deviation than in women. In addition, sex hormones may influence ischemic susceptibility, with androgens promoting vasoconstriction and estrogens favoring vasodilation48. In women, age-related changes in sex hormone levels, particularly the decline in estrogen after menopause, may modify vascular function, hypoxia susceptibility, and erythropoietic regulation. Although detailed hormonal or menopausal status data were not available in this cohort, this limitation may partially influence the observed associations between anemia and renal prognosis. Future studies incorporating menopausal status and/or hormonal measurements are warranted to clarify sex-specific mechanisms. Collectively, these factors may help explain why anemia appears to convey a stronger prognostic signal in men with ADPKD20.

Current CKD guidelines recommend anemia treatment at relatively low hemoglobin thresholds49,50,51,52,53,54,55; however, whether these targets are optimal for patients with ADPKD, particularly men, remains uncertain. Our findings suggest that anemia in ADPKD may function primarily as a marker of underlying disease severity rather than a direct therapeutic target. Sex-related differences in metabolic demand and hypoxia susceptibility may provide a possible physiological context for the observed associations.

This study has several limitations. First, its observational design precludes causal inference. Second, analyses were based on baseline measurements and did not account for longitudinal changes in hemoglobin or other covariates during follow-up. In addition, the time-series pseudo-R² analyses were intended to explore the temporal consistency of prognostic associations rather than to provide individual-level risk prediction. Accordingly, these results should be regarded as hypothesis-generating, and external validation in independent ADPKD cohorts is warranted.

In conclusion, lower hemoglobin levels were modestly but significantly associated with poorer long-term renal outcomes in patients with ADPKD. Using a time-series pseudo-R² framework, we compared the relative contribution of prespecified Hb cut-offs over follow-up time and observed that candidate thresholds—approximately 12 g/dL overall, 13 g/dL in men, and 12 g/dL in women—showed more persistent associations with renal prognosis. These findings should be interpreted as hypothesis-generating and complementary to conventional survival analyses. Rather than indicating a direct therapeutic target, anemia may reflect underlying disease severity and progression, particularly in men. Further external validation and mechanistic studies are warranted before any clinical translation or interventional implications can be considered.