Charlson comorbidity index and palliative performance scale predict prognosis in dialysis patients: a retrospective cohort study

Abstract

Background

Accurate prognostication in end-stage kidney disease (ESKD) is essential for informed clinical decision-making, timely palliative care integration, and aligning treatments with patient goals. Commonly used tools include the Charlson Comorbidity Index (CCI) and the Palliative Performance Scale (PPS), although their comparative predictive performance in dialysis populations remains unclear. This study aimed to evaluate and compare the prognostic accuracy of the CCI, pre-dialysis PPS, and post-dialysis PPS in predicting natural mortality among patients with ESKD receiving renal replacement therapy (RRT).

Methods

This retrospective cohort study included 1,047 adult patients with ESKD who initiated hemodialysis or peritoneal dialysis at a tertiary hospital in southern Thailand between 2009 and 2022. The CCI was calculated based on diagnoses within 90 days prior to dialysis. PPS was assessed within 30 days before (pre-dialysis) and after (post-dialysis) initiation. The primary outcome was natural mortality. Predictive performance was evaluated using Kaplan–Meier survival analysis, multivariable Cox regression, and time-dependent receiver operating characteristic (ROC) curves at 30, 180, and 365 days.

Results

During a median follow-up of 5.89 years, 58.0% of patients died. The CCI effectively stratified long-term survival (median survival: 11.27 years for score < 6 vs. 3.80 years for > 8). However, for short-term prognosis, the post-dialysis PPS demonstrated superior discriminative ability. At 30 days, the continuous post-dialysis PPS model achieved a time-dependent AUC of 0.91 (95% CI 0.85–0.99), significantly outperforming the CCI (AUC 0.76). In multivariable analysis, age and post-dialysis PPS were associated with mortality. Functional recovery following dialysis initiation emerged as a critical marker of physiological resilience.

Conclusions

The CCI and PPS offer complementary prognostic value in ESKD. While the CCI reflects the accumulated burden of disease affecting long-term survival, the post-dialysis PPS captures dynamic physiological resilience and is a superior predictor of early mortality. Incorporating post-dialysis functional assessment into routine care may enhance risk stratification, identifying high-risk patients who would benefit from early palliative care integration and goals-of-care discussions.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12904-026-01989-2.

Background

End-stage kidney disease (ESKD) represents a significant global health burden, with a rising prevalence and substantial morbidity and mortality despite advances in dialysis technology. Although renal replacement therapy (RRT) prolongs the life of many patients, outcomes vary widely depending on patient-specific factors such as comorbidities and functional status [1, 2]. Therefore, accurate prognostication of ESKD is critical for informing clinical decision-making, optimizing the timing of palliative care integration, and supporting shared decision-making regarding dialysis initiation and continuation [3].

The Charlson Comorbidity Index (CCI) and performance status are two widely used prognostic tools in chronically ill populations. The CCI is a validated instrument that quantifies comorbidity burden and has been adapted for use in ESKD populations [4, 5]. Higher CCI scores have been consistently associated with increased mortality in patients undergoing hemodialysis and peritoneal dialysis patients [6, 7, 8].

For functional assessment, the Karnofsky Performance Status Scale (KPS) has been historically used in nephrology and oncology populations [9]. However, KPS primarily evaluates activity and independence and may be less sensitive in capturing the multidimensional decline seen in palliative and advanced organ failure patients [10]. In contrast, the Palliative Performance Scale (PPS) has demonstrated stronger prognostic discrimination in palliative cohorts, with higher C-index values compared to KPS, supporting its use as a more reliable functional measure in this context [11].

PPS adapted from the KPS, was developed for palliative care and incorporates additional domains such as oral intake and level of consciousness [12]. These domains are particularly relevant in ESKD, where anorexia, reduced intake, and fluctuating consciousness often accompany disease progression and treatment intolerance. In Thailand, the Suandok PPS is a direct Thai-language translation of the original PPS and has been routinely applied in palliative care practice [13].

PPS has demonstrated prognostic value beyond cancer populations, including patients with chronic disease as advanced kidney disease [10, 11, 14]. Compared to KPS, PPS provides a more nuanced picture of decline and survival trajectories in patients with palliative care needs.

Despite its increasing use, the comparative prognostic value of these tools for ESKD, particularly when the PPS is assessed pre- and post- dialysis initiation, remains insufficiently explored. Understanding the interplay of comorbidity burden and functional status may offer a more comprehensive view of patient trajectories and help tailor care plans according to individual needs [15, 16]. In particular, changes in functional status after dialysis initiation may reflect early signs of treatment intolerance or limited benefits, particularly among older adults with a high comorbidity burden [17].

In this study, we aimed to perform a head-to-head comparison of the prognostic accuracy of the CCI against both pre-dialysis and post-dialysis PPS in a large Thai cohort of patients with ESKD receiving RRT. Beyond simple validation, this study specifically sought to evaluate the functional trajectory surrounding the initiation of therapy. By examining these tools in parallel, we aimed to determine whether functional recovery after dialysis initiation (measured via post-dialysis PPS) offers complementary prognostic value to established comorbidity-based models, thereby supporting clinical decision-making and the timely integration of palliative care.

Methods

Study design and setting

This retrospective cohort study was conducted at the Songklanagarind Hospital, a tertiary care academic center in southern Thailand. The study period spanned January 2009 to December 2022. The year 2009 was chosen as the starting point because it corresponds to the full implementation and comprehensive availability of our institution’s electronic health record (EHR) system. This ensured that we could reliably capture the required data for the study variables, including diagnoses and functional assessments, from that point forward.

Study population

The study included adult patients (aged ≥ 18 years) with a confirmed diagnosis of ESKD who initiated either hemodialysis or peritoneal dialysis and continued therapy for at least 2 months. The exclusion criteria were as follows: (1) death from unnatural causes including homicide, suicide, accident and undetermined, (2) receipt of kidney transplantation, (3) transferred to another facility before the post-dialysis PPS evaluation (conducted within 30 days after initiation), and (4) incomplete medical records, defined as medical record with fewer than 1,000 characters, which was used as a data quality proxy indicating that the record lacked sufficient clinical detail to accurately calculate the CCI or PPS. Patients were identified from hospital records using the diagnostic and procedural codes for ESKD and dialysis.

Prognostic measures

Two primary prognostic tools were assessed. First, the CCI assigns weighted scores to 19 comorbid conditions, with additional age-based points awarded to patients aged > 40 years (one point per decade). CCI scores were calculated based on diagnoses documented within 90 days before dialysis initiation [4]. Second, the PPS evaluates functional status across five domains: ambulation, activity level, self-care, oral intake, and level of consciousness. The scores range from 0% (death) to 100% (fully functional) [12]. The validated Thai version of the Suandok scale was used in this study. PPS scores were recorded at two timepoints: within 30 days before dialysis initiation (pre-dialysis PPS) and within 30 days after dialysis initiation (post-dialysis PPS) [13]. This specific 30-day interval was selected to capture the immediate functional trajectory surrounding the initiation of therapy; the pre-dialysis score established a baseline, while the post-dialysis score was intended to assess early functional recovery and physiological resilience following the acute transition from renal replacement therapy.

Outcome measures

The primary outcome measure was the natural mortality. Survival time was calculated from the date of dialysis initiation to the date of death. For patients who remained alive or were lost to follow-up, data were censored at the date of their last hospital visit or December 31, 2022, whichever came first. Mortality data were verified using both hospital records and the National death registry of the Division of Data and Innovation Analytics and Ministry of Public Health, Thailand, respectively.

Statistical analysis

Descriptive statistics were used to summarize baseline characteristics and are presented as medians with interquartile range (IQR) for continuous variables and frequencies with percentages for categorical variables. Survival probabilities were estimated using Kaplan–Meier curves, and differences between groups were assessed using log-rank tests. Univariate and multivariate cox proportional hazards regression models were employed to evaluate independent associations between prognostic factors and survival outcomes. Variables yielding the highest concordance index, along with clinically relevant demographic variables (age, sex, and dialysis modality), were entered into the multivariate model using a forced-entry method. Results are presented as hazard ratios (HR) with 95% confidence intervals (CI). Predictive performance was assessed using the time-dependent area under the receiver operating characteristic curve (AUC) and the concordance index (C-index). Static models often inflate performance estimates by analyzing fixed outcomes and excluding censored data. In contrast, time-dependent analysis rigorously accounts for censoring and variable follow-up, typically yielding lower, more conservative AUC estimates that accurately reflect the natural “dilution” of prognostic signal over extended periods [18]. To evaluate the incremental predictive value of functional status following dialysis initiation, we compared the C-indices across models and analyzed time-dependent diagnostic accuracy at prespecified horizons of 30 days, 180 days (6 months), and 365 days (1 year). For each time horizon, accuracy metrics—including sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and AUC—were calculated along with their 95% CI (Supplementary Tables 1 and 3).

To facilitate clinical interpretation and risk stratification, the prognostic scores were categorized into distinct groups based on the distribution of data and clinical relevance. For the group models, the CCI was stratified into three levels of comorbidity burden groups (< 6, 6–8, and > 8). The PPS was analyzed as pre-defined groups (0–30, 40–70, and 80–100) to explore potential non-linear relationships with survival. For the continuous models, diagnostic metrics were derived using the standard probability threshold of 0.5 for the purpose of model comparison. All analyses were conducted using R Version 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria).

Ethical considerations and consent to participate

This study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Human Research Ethics Committee of the Faculty of Medicine, Prince of Songkla University (REC.67–366-9-4). Patient consent was waived due to the retrospective nature of the study and the use of de-identified data.

Results

A total of 1,047 patients with ESKD receiving dialysis were included in the analysis. Of these, 440 patients (42.0%) were alive at the end of the follow-up period, while 607 (58.0%) had died. Baseline characteristics stratified by survival status are presented in Table 1.

Table 1.

Patient characteristics by survival status

Variable	Alive (n = 440)	Death (n = 607)	Total (N = 1047)	P-value
Age, median (IQR), years	59.2 (45.7–68.1)	67.7 (57.4–75.7)	63.8 (53.4–73.0)	< 0.001
Age group, n (%)				< 0.001
0–39	79 (18.0)	29 (4.8)	108 (10.3)
40–49	63 (14.3)	38 (6.3)	101 (9.6)
50–59	86 (19.5)	119 (19.6)	205 (19.6)
60–69	122 (27.7)	162 (26.7)	284 (27.1)
70–79	66 (15.0)	180 (29.7)	246 (23.5)
80–89	24 (5.5)	70 (11.5)	94 (9.0)
90–99	0 (0.0)	9 (1.5)	9 (0.9)
Sex, n (%)				0.680
Male	245 (55.7)	329 (54.2)	574 (54.8)
Female	195 (44.3)	278 (45.8)	473 (45.2)
Weight, median (IQR), kg	62.0 (54.0–72.2)	60.0 (52.3–70.0)	61.0 (53.0–71.0)	0.006
Height, median (IQR), cm	160 (155–167)	160 (154–165)	160 (155–166)	0.324
Religion, n (%)				0.489
Buddhism	376 (85.6)	534 (88.0)	910 (87.0)
Islam	62 (14.1)	71 (11.7)	133 (12.7)
Christianity	1 (0.2)	2 (0.3)	3 (0.3)
Type of RRT, n (%)				0.005
Hemodialysis	382 (86.8)	486 (80.1)	868 (82.9)
Peritoneal dialysis	58 (13.2)	121 (19.9)	179 (17.1)
CCI, median (IQR)	6 (4–8)	8 (6–9)	7 (5–9)	< 0.001
CCI group, n (%)				< 0.001
< 6	198 (45.0)	123 (20.3)	321 (30.7)
6–8	161 (36.6)	271 (44.6)	432 (41.3)
> 8	81 (18.4)	213 (35.1)	294 (28.1)
Pre-dialysis PPS, n (%)				< 0.001
0–30	7 (1.7)	18 (3.2)	25 (2.6)
40–70	50 (12.2)	139 (25.1)	189 (19.6)
80–100	354 (86.1)	397 (71.7)	751 (77.8)
Post-dialysis PPS, n (%)				< 0.001
0–30	2 (0.5)	35 (6.3)	37 (3.8)
40–70	20 (4.8)	81 (14.5)	101 (10.4)
80–100	394 (94.7)	443 (79.2)	837 (85.8)

IQR Interquartile range, RRT Renal replacement therapy, CCI Charlson Comorbidity Index, PPS Palliative Performance Scale

The median age was significantly higher among deceased patients compared to survivors (67.7 years [IQR 57.4–75.7] vs. 59.2 years [IQR 45.7–68.1]; P < 0.001). Age distribution also differed notably, with a greater proportion of patients aged ≥ 70 years in the deceased group (42.7%) than in the survivor group (20.5%). Sex distribution was similar between groups (male: 55.7% alive vs. 54.2% deceased; P = 0.68).

Deceased patients had a significantly lower median body weight than survivors (60.0 kg [IQR 52.3–70.0] vs. 62.0 kg [IQR 54.0–72.2]; P = 0.006). Median height did not differ significantly between the groups (P = 0.324). Most patients in both groups practiced Buddhism, with no significant difference in religious affiliation (P = 0.489).

Dialysis modality varied significantly by survival status. Peritoneal dialysis was more common among deceased patients (19.9% vs. 13.2%; P = 0.005), whereas hemodialysis was more frequent among survivors (86.8% vs. 80.1%).

Comorbidity burden, as assessed by the CCI, was significantly higher in the deceased group (median 8 [IQR 6–9]) than in the survivor group (median 6 [IQR 4–8]; P < 0.001). A CCI > 8 was observed in 35.1% of deceased patients and 18.4% of survivors.

Functional status, measured by the PPS, was also significantly worse among deceased patients, both before and after dialysis initiation. Prior to dialysis, 25.1% of deceased patients had a PPS of 40–70 and 3.2% had a PPS of 0–30, compared to 86.1% of survivors who had a PPS of 80–100 (P < 0.001). After dialysis initiation, 20.8% of deceased patients had a PPS < 80, compared to only 5.3% of survivors (P < 0.001).

The overall median survival time for the entire cohort was 5.89 years (95% CI 5.13–6.51). Stratified survival outcomes are summarized in Table 2. Survival varied significantly by CCI score. Patients with a CCI < 6 had the longest median survival (11.27 years; 95% CI 9.80–NA), followed by those with a CCI of 6–8 (5.17 years; 95% CI 4.36–6.10). Patients with a CCI > 8 had the shortest survival (3.80 years; 95% CI 3.29–4.22).

Table 2.

Median survival time by patient group

Patient Group	Number of Patients (n)	Number of Events	Median Survival, years	95% Confidence Interval
Overall	1047	607	5.89	5.13–6.51
CCI
Score < 6	321	123	11.27	9.80–NA
Score 6–8	432	271	5.17	4.36–6.10
Score > 8	294	213	3.80	3.29–4.22
Pre-dialysis PPS
Score 0–30	25	18	0.71	0.45–NA
Score 40–70	189	139	3.08	2.40–3.60
Score 80–100	751	397	6.36	5.57–7.12
Post-dialysis PPS
Score 0–30	37	35	0.43	0.28–0.84
Score 40–70	101	81	2.34	1.68–3.01
Score 80–100	837	443	6.75	6.09–7.13

CCI Charlson Comorbidity Index, PPS Palliative Performance Scale

Pre-dialysis PPS was also prognostically significant. Patients with a PPS of 80–100 had a median survival of 6.36 years (95% CI 5.57–7.12), compared to 3.08 years (95% CI 2.40–3.60) for those with a PPS of 40–70 and 0.71 years (95% CI 0.45–NA) for those with a PPS of 0–30.

Post-dialysis PPS showed similar prognostic value. Median survival among patients with a Post-dialysis PPS of 80–100 was 6.75 years (95% CI 6.09–7.13), compared to 2.34 years (95% CI 1.68–3.01) for PPS 40–70 and 0.43 years (95% CI 0.28–0.84) for PPS 0–30.

Kaplan–Meier survival curves stratified by CCI score are shown in Fig. 1. Patients with a CCI < 6 demonstrated the highest survival probabilities throughout the follow-up period. The curves diverged early and remained separated over time, with significant differences confirmed by the log-rank test (P < 0.001).

Fig. 1.

Kaplan–Meier survival curves stratified by Charlson Comorbidity Index groups

Figure 2 presents Kaplan–Meier survival curves by pre-dialysis PPS. Patients with PPS scores of 80–100 had markedly better survival than those with lower scores, with the steepest decline observed in the PPS 0–30 group (P < 0.001).

Fig. 2.

Kaplan–Meier survival curves stratified by pre-dialysis Palliative Performance Scale scores

Similarly, Kaplan–Meier analysis stratified by post-dialysis PPS (Fig. 3) showed that higher scores following dialysis initiation were associated with significantly improved survival. The survival curves remained distinctly separated across PPS categories throughout follow-up (P < 0.001).

Fig. 3.

Kaplan–Meier survival curves stratified by post-dialysis Palliative Performance Scale scores

The multivariate cox regression analysis (Table 3 and Supplementary Table 1) identified four factors associated with mortality. Age and post-dialysis PPS were the strongest factors associated with mortality. For every 1-year increase in age, the hazard of death increased by 3% (adjusted Hazard Ratio (aHR) = 1.03), while every 1-point increase in the Post-dialysis PPS score (indicating better functional status) resulted in a 3% decrease in the hazard of death (aHR = 0.97). Furthermore, demographic and treatment factors also played a significant role. Being male increased the hazard of death by 24% compared to females (aHR = 1.24), and receiving peritoneal dialysis increased the hazard of death by 38% compared to hemodialysis (aHR = 1.38).

Table 3.

Multivariable Cox regression analysis of factors associated with mortality

Characteristics	Univariate HR (95% CI)	Univariate p-value	Multivariate HR (95% CI)	Multivariate p-value
Post-dialysis PPS	0.97 (0.96–0.97)	< 0.001	0.97 (0.97–0.98)	< 0.001
Age (Per 1 year increase)	1.04 (1.03–1.05)	< 0.001	1.03 (1.02–1.04)	< 0.001
Sex
Female	—		—
Male	1.23 (1.01–1.48)	0.036	1.24 (1.02–1.50)	0.028
Type of RRT
Hemodialysis	—		—
Peritoneal Dialysis	1.25 (0.98–1.58)	0.070	1.38 (1.08–1.76)	0.009

CI Confidence Interval, HR Hazard Ratio, PPS Palliative Performance Scale

RRT Renal Replacement Therapy

The post-dialysis PPS model demonstrated significantly higher overall discriminative ability (C-index = 0.69; 95% CI 0.66–0.71) compared to both the pre-dialysis PPS model (C-index = 0.59; 95% CI 0.56–0.62) and the CCI model (C-index = 0.62; 95% CI 0.59–0.65). Furthermore, the multivariable model incorporating Post-dialysis PPS (Model 5) yielded the highest overall predictive accuracy (C-index = 0.70; 95% CI 0.67–0.73) (Supplementary Table 1).

The time-dependent analysis of diagnostic accuracy revealed that models incorporating the post-dialysis PPS consistently outperformed those based solely on the CCI, particularly in the short-to-medium term (at 30 and 180 days). The continuous post-dialysis PPS only model demonstrated the highest discriminative ability, achieving an AUC of 0.91 (95% CI 0.85–0.99) at 30 days, compared to 0.76 (95% CI 0.63–0.86) for the continuous CCI model. This performance advantage persisted at 6 months, where the continuous PPS model maintained an AUC of 0.85 versus 0.64 for the CCI. Multivariable models that adjusted for age, sex, and dialysis modality further validated these findings, maintaining robust predictive power with AUCs ranging from 0.78 to 0.87 across the 30- and 180-day horizons (Fig. 4 and Supplementary Table 3).

Fig. 4.

Time-Dependent Area Under the ROC Curve (AUC) for Continuous (A) and Grouped (B) Prognostic Models: The plots display the dynamic discriminative ability (AUROC) of various prognostic models over the first year (365 days) following dialysis initiation. Figure 4A illustrates the performance of models using continuous score variables, while Figure 4B illustrates models using stratified risk groups

When analyzed as stratified risk groups, distinct performance profiles emerged that favor the utility of the PPS for identifying high-risk patients. While the CCI Group model exhibited perfect sensitivity (100%) at 30 days, it was severely limited by low specificity (28.7%), indicating a high rate of false positives where many survivors were classified as high-risk. In contrast, the post-dialysis PPS group models offered a superior balance, demonstrating high specificity (ranging from 89.4% to 97.8%) and a consistently high NPV (> 96%) across all time points (Fig. 4 and Supplementary Table 3).

Discussion

This study confirms that Post-dialysis PPS and Multivariable models demonstrate superior predictive accuracy compared to the standard CCI, particularly in the short-to-medium term. The Continuous Post-dialysis PPS model achieved the highest discriminative ability (AUC 0.91 at 30 days; 0.85 at 180 days), significantly outperforming the CCI (AUC 0.76 and 0.64, respectively). This indicates that early functional status is a more sensitive mortality indicator than comorbidity burden alone.

Distinct performance characteristics were observed among the group models. While the CCI group model offered high sensitivity (100%) at the expense of very low specificity (28.7%), the post-dialysis PPS models provided a balanced profile with high specificity (> 97%) and superior PPV. This makes PPS models more effective for identifying high-risk patients suitable for palliative care, whereas the PPS Group model’s high NPV (NPV > 96%) effectively rules out low-risk patients.

These findings extend prior CCI validation by highlighting early functional recovery as a key marker of physiological resilience. Furthermore, our early predictive power (AUC > 0.85) exceeds that of similar studies (AUC 0.70–0.80), noting that our use of time-dependent ROC analysis provides rigorous, conservative estimates by accounting for censored data over long follow-up periods [11, 19]. This finding aligns with prior research by Prompantakorn et al. In their cohort, the PPS showed strong discriminative ability for cancer patients and non-cancer groups [19]. In the specific context of renal failure, functional decline is often more gradual and interspersed with acute episodes compared to the steady decline observed in malignancies. This may explain why our long-term AUC values (0.58–0.69 at 1 year) are more modest than those typically reported in cancer-specific palliative cohorts. However, our study demonstrates that in the short term (30 days), the discriminative ability of the post-dialysis PPS (AUC > 0.90) is comparable to or exceeds that of prognostic models in other advanced chronic conditions. For instance, Affizal et al. reported AUCs ranging from 0.70 to 0.90 for the palliative prognostic indexes including PPS and CCI in geriatric populations [20]. Forzley et al. reported similar values for a mortality prediction model at the C-index of 0.70 based on “the surprise question” [21].

While the Karnofsky Performance Status Scale (KPS) is historically used, its utility in ESKD may be limited as it primarily assesses activity and independence. PPS was selected for evaluation in this study because it adds domains highly relevant to ESKD, such as oral intake and level of consciousness, which can reflect severe disease progression like anorexia. Prior studies also suggest the PPS has stronger prognostic discrimination than KPS in palliative cohorts. The PPS demonstrated a clear prognostic value in this study, with lower scores associated with significantly shorter survival. This finding aligns with prior research in both cancer and non-cancer populations, where the PPS has been shown to correlate with mortality risk and inform palliative care needs [11, 14]. More recently, evidence has extended its utility to high-risk non-oncology groups, including patients with advanced kidney disease, where the PPS can aid in identifying patients who may benefit from early palliative interventions [22]. A key contribution of this study is the finding that post-dialysis PPS demonstrates superior prognostic accuracy compared with pre-dialysis PPS. This suggests that early functional recovery following dialysis initiation may serve as a surrogate marker of physiological resilience, offering valuable insights into patient prognosis and potential responsiveness to ongoing treatment.

Multivariate cox regression analysis identifies age and post-dialysis PPS as the most robust factors associated with mortality. Advanced age increased the hazard of death, while better functional status—indicated by a higher PPS—conferred a better survival, underscoring functional reserve as a critical prognosticator of comorbidity burden [23]. Notably, peritoneal dialysis had a higher hazard of death compared to hemodialysis. This disparity likely reflects the selection bias inherent in Thailand’s “PD First” policy, where modality choice is often dictated by insurance mandates rather than clinical optimization, consistent with regional findings of elevated PD mortality risks in specific Asian subgroups [24].

All the prognostic tools evaluated in this study demonstrated only modest discriminative ability. This finding underscores the multidimensional nature of prognostication in advanced kidney disease, reinforcing prior evidence that no single measure can fully capture the diverse clinical, functional, and psychosocial determinants of survival [25]. These tools should not be used as standalone, definitive predictors. Instead, their primary clinical utility lies in their role as complementary instruments to support a holistic clinical assessment, help identify high-risk patients (e.g., CCI > 8 or PPS < 70), and serve as a trigger for goals-of-care discussions and timely palliative care integration.

Clinical implications

The CCI and the post-dialysis PPS offer clinicians a practical and evidence-based strategy for risk stratification. Our findings highlight that these tools are complementary. While the CCI reflects the accumulated burden of disease, the post-dialysis PPS captures the patient’s dynamic physiological response to the stress of initiating renal replacement therapy. Consequently, functional recovery serves as a critical prognostic indicator that complements comorbidity-based models.

While the predictive accuracy for an individual is modest, the clinical utility lies in identifying high-risk groups—specifically those with a high comorbidity burden (CCI > 8) or an inability to recover functional status (PPS < 70)—who would most benefit from early, proactive palliative care integration and shared decision-making. Recognizing this subgroup early in the treatment course enables more proactive and personalized care planning. This includes the timely integration of palliative care services, facilitation of advance care planning discussions, and shared decision-making regarding the continuation or withdrawal of dialysis. Incorporating these prognostic tools into routine nephrology workflows—not as standalone predictors but as part of a holistic assessment—has the potential to improve the alignment of treatment strategies with patient values and goals, reduce unnecessary interventions, and enhance the quality of end-of-life care [17, 18, 26].

Strengths and limitations

This study has several notable strengths. It includes a large, well-defined cohort of patients with ESKD, and the use of validated prognostic tools (CCI and PPS). Importantly, the inclusion of both pre- and post-dialysis PPS assessments provides a unique perspective on the functional trajectory following the initiation of renal replacement therapy, highlighting its dynamic nature and potential prognostic relevance.

However, this study has some limitations. First, reliance on baseline scores to predict long-term outcomes ignores longitudinal changes; while we analyzed 6- and 12-month intervals, the predictive value of baseline scores naturally diminishes over time (“dilution effect”). Second, the retrospective, single-center design may limit the generalizability of the results. Third, residual confounding may arise from unmeasured variables (e.g., frailty, biomarkers). Fourth, historically the choice of dialysis modality (hemodialysis vs. peritoneal dialysis) in Thailand is often heavily influenced by national health insurance policies rather than patient preference, which may act as a potential confounder. Fifth, the focus on mortality excludes broader patient-centered outcomes. Sixth, excluding incomplete records creates potential selection bias, possibly underestimating mortality. Seventh, retrospective PPS scoring relies on documentation quality, carrying a risk of measurement bias compared to prospective assessment. Finally, immortal time bias is possible, as the requirement to survive until assessment excludes patients with the most rapid mortality.

Suggestions for future research

Future research should investigate whether combining comorbidity indices such as the CCI with dynamic assessments of functional status (e.g., serial PPS measurements), frailty indices, and symptom burden can improve prognostic model accuracy in ESKD. Prospective, multicenter studies are needed to validate these integrated models across diverse clinical settings. Furthermore, as this retrospective study was limited by the lack of systematically recorded KPSS data, future prospective research would benefit from a direct comparison of the prognostic accuracy of the PPS and the KPSS in the ESKD population. Additionally, research should explore whether employing these tools to guide shared decision-making influences clinical outcomes, improves the timing and quality of palliative care integration, and enhances alignment between treatment decisions and patient preferences.

Conclusion

In patients with ESKD, the CCI robustly predicts long-term outcomes, while post-dialysis PPS offers superior short-term value at 30 days, identifying early functional recovery as a key marker of resilience. CCI and PPS may play complementary roles; however, these findings—particularly regarding long-term prediction—are exploratory and require prospective validation. Clinical application should proceed with caution, utilizing these tools not as standalone determinants, but to identify high-risk patients and guide shared decision-making.

Authors’ contributions

YR contributed to study conception, data collection, and drafting of the manuscript. PL performed statistical analyses and interpretation of results. OF supervised the project, critically revised the manuscript, and was the corresponding author. PS contributed to patient recruitment and clinical data verification. DT performed data management and supported statistical modeling. TI provided methodological guidance, supervised data analysis, and contributed to manuscript revision. All authors read and approved the final manuscript.

Funding

This research was conducted independently and received no funding from any source.

Data availability

The datasets generated and/or analyzed during the current study are not publicly available due to institutional restrictions and patient confidentiality, but de-identified data may be available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Human Research Ethics Committee of the Faculty of Medicine, Prince of Songkla University (REC.67-366-9-4). Patient consent was waived due to the retrospective nature of the study and the use of de-identified data.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.