Data source and methods of data collection

The Chronic Kidney Disease in Queensland (CKD.QLD) was established as a collaborative of most public sector kidney care practices in Queensland in May 2011. The CKD.QLD data collection methods have been described by Venuthurapalli et al, briefly however, the main objective of the registry was to profile participants with CKD, promoting CKD surveillance, practice improvement and research within the kidney care practice network in the public health system in QLD [19]. Data linkage framework was used to centralise data captured by multiple mechanisms to an individual participant via a unique identifier. This enabled further characterisation of participants with CKD, providing a comprehensive view of the health services utilisation and costs and outcomes of people with CKD.

Adult participants aged 18 years and above with the diagnosis of CKD presenting to public ambulatory nephrology clinics in the QLD health (QH) system were consented to have their demographic and clinical data collected. More than 7,600 participants consented to be part of the Registry between June 2011 and June 2018. Excluded were patients on KRT and those with acute kidney injury (AKI) unless they subsequently developed and met the diagnostic criteria for CKD. Participants were followed until the start of KRT, death or 30 June 2018. For the current study, we excluded participants whose first date of visit to the nephrology clinic couldn’t be established.

The demographic variables recorded included age, gender, ethnicity and geographical location, identified by postcode, as well as laboratory and clinical variables and outcome measures. Variables of interest which were not already in the database were manually collected from integrated electronic Medical Record (ieMR), the Hospital Based Corporate Information System (HBCIS), other QLD health platforms, and from private pathology laboratories as appropriate, following granting of a Public Health Act (PHA) approval. Where only urine protein-to-creatinine ratio (uPCR) and dipstick protein were available, the equations developed by Sumida et al were employed to calculate the predicted urine albumin-to-creatinine ratio (UACR) [20]. The eGFR was calculated from serum creatinine using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation. Data extraction was undertaken between 18 January 2021 and 15 June 2025 in compliance with ethics and PHA approvals.

Study design and population

This retrospective cohort study utilised data from participants aged ≥18 years enrolled in the CKD.QLD Registry between 1 May 2011 and 30 June 2018 from seven public healthcare nephrology clinics. A description of participants and basic study outcomes were previously detailed in a published protocol [21].

Outcomes and variables

Primary outcomes included progression to end-stage kidney disease (ESKD), commencement of KRT, and mortality before and after commencement of KRT. Demographic and clinical variables were investigated for their potential role as predictors or covariates of these outcomes. Referral timing to nephrology services was assessed and defined by the conventional method as the interval between the first encounter with the nephrology service and commencement of KRT. Late referral was defined as initiation of KRT within 12 months of the initial specialist nephrology referral among patients who progressed to KRT. This threshold reflects the time typically required for clinical optimisation, patient education, and establishment of permanent vascular access [22]. To assess the robustness of this definition, a sensitivity analysis was performed using a multivariable left-truncated Fine-Gray model, with referral duration modelled continuously and compared at 1, 3, 6, and 12 months of pre-KRT care (Figure S3 in S1 File).

The proportion of participants meeting the KHA eGFR referral threshold (<30 mL/min) at referral was summarised descriptively. This threshold was selected over other criteria (e.g., eGFR decline or microscopic haematuria) as eGFR-based cut-offs have traditionally guided key CKD management decisions, including specialist referral, access planning, and transplant evaluation, and because urine microscopy data were unavailable for most participants [23]. The Kidney Failure Risk Equation (KFRE) was applied to participants with CKD stages G3-G5 at referral with available UACR data to estimate 5-year risk of progression to KRT. KFRE risk distributions were summarised descriptively to explore alignment across risk thresholds. Model discrimination and calibration were not assessed.

Socio-Economic Indexes for Areas (SEIFA) Relative Socio-economic Disadvantage (IRSD) scores as produced by the Australian Bureau of Statistics were used to measure socioeconomic status of the enrolled participants by their residential postcode [24]. The IRSD deciles were regrouped into quintiles (lowest, low, middle, high, and highest) with the highest quintile representing the least disadvantaged postal areas and the lowest quintile representing the most disadvantaged postal areas [25]. Participants were further categorised into 3 levels of socio-economic status by combining the two lowest quintiles into the low socio-economic status group, the two highest quintiles into the high socio-economic status group, and the middle quintile into the middle socio-economic status group. Participants with a recorded BMI were divided into obese (BMI ≥ 30) and non-obese (BMI < 30). A modified and unweighted version Charlson Comorbidity Score (also referred to as a simple count of comorbid conditions) was used to predict mortality risk. The weighted Charlson Comorbidity Index (CCI) could not be used because the registry did not contain sufficiently detailed information on the individual conditions required to calculate the score. While the original CCI assigns weights (1–6) to different conditions, research indicates that simply counting the number of conditions often provides similar predictive power for long-term mortality [26]. The CCI scores were classified into four groups based on the distribution in the study population: low = 2 (scored by ESKD per se), moderate = 3–4, high = 5–6, and very high > 6 [27].

Data analysis

Descriptive statistics and basic inferential statistics were employed to summarise demographic and clinical characteristics, exploring basic data patterns and describing referral patterns. Categorical variables were presented as proportions (numbers and percentages) and continuous variables as means with standard deviations.

For the primary analysis, Cox proportional hazards regression was used. This analysis was restricted to participants who initiated KRT (n = 513), with time zero defined as the date of KRT initiation; deaths before KRT were excluded. Covariates for the full pre-specified model were chosen based on prior clinical knowledge of mortality risk in ESKD, and pre-specified confounder set was retained regardless of univariate significance [28]. The proportional hazards assumption was assessed using Schoenfeld residuals and no significant violations were observed for the overall model (p = 0.79) or for individual covariates. Results are presented as hazard ratios (HRs) with 95% confidence intervals (CIs). Kaplan-Meier curves and log-rank tests were used to compare unadjusted survival between early and late referral groups.

To address potential biases from restricting analysis to patients who reached KRT, a complementary competing risks analysis was conducted in the full cohort (n = 3,775). A Fine-Gray subdistribution hazards model with left-truncation (delayed entry) was applied. Time zero was defined as the date of initial nephrology referral, allowing deaths before KRT (n = 598) to be included as competing events. Patients entered the risk set at the time of KRT initiation and followed up until the end of the study period (post-KRT deaths = 124; censored = 389). Statistically, truncation reduces instability from diminishing risk sets at longer durations and limits potential violations of proportional hazards assumptions. It also improves visual clarity by preserving early event dynamics and enhancing interpretability of cumulative incidence functions. The multivariable model included duration of pre-KRT care, age, sex, Indigenous status, social status score, comorbidity score, and histories of DM, hypertension, CVD, and coronary artery disease. The proportional hazards assumption for the subdistribution model was assessed using time-varying covariate interactions with log time. Adjusted cumulative incidence functions (CIF) were generated using stcurve to compare post-KRT mortality across referral durations of 1, 3, 6, and 12 months (Figure S3 in S1 File). All analyses were performed using Stata (version 18/19), with statistical significance defined as p < 0.05.

Ethical considerations

The CKD.QLD registry and the hospital record(s) were examined by the research team retrospective to the participants’ details being entered into CKD.QLD and their hospital admissions. Both CKD.QLD data and hospital records contained identifiable information essential for data linkage. At enrolment, participants provided informed consent for CKD.QLD, which included permission to access and link relevant clinical information, such as medical history, pathology reports, and prior hospital admissions. Due to the large number of participants, obtaining individual consent for this study was not feasible. Instead, researchers obtained approval from the CKD.QLD Data Custodian to use existing consent arrangements under HREC/15/QRBW/294 (ERM: 13429) (AM05). Ethics approval was granted by the Royal Brisbane and Women’s Hospital Human Research Ethics Committee in January 2021 (LNR/2020/QRBW/69707, 14/01/2021) to access hospital records using unit record numbers provided by CKD.QLD for matching, with the intention to publish the findings. The requirement for individual consent was waived by the Ethics Committee, as all participants had previously consented to the use of their data for research upon registry enrolment. Additionally, a Public Health Act approval (QCOS/029817/ RD006802) was granted in August 2023 for the release of missing patient data held by the contributing Queensland Health Hospital and Health Services.

Study sample characteristics

Figure S1 in S1 File outlines the participant selection process. A total of 3,775 participants were included in the final analysis with mean age of 62.5 ± 15.1 years. Of these, 2,069 (54.8%) were male and 327 (8.7%) identified as Indigenous Australians (Table 1).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Baseline characteristics at the time of referral and outcomes.

https://doi.org/10.1371/journal.pone.0338001.t001

Socioeconomic status, as measured by the IRSD, revealed that a significant portion of the participants (1,806, 47.8%) lived in the most disadvantaged postal areas (low socioeconomic status) (Table 1). This group represented the largest proportion of participants across all stages of CKD (Table 1).

At the time of referral, nearly half of the participants, 1,872 (49.6%) were in stage G3 CKD (G3A and G3B), whereas 713 (18.9%) were in stages 1 or 2 (eGFR of ≥60 ml/min/1.73m²). CKD stages G4 and 5 accounted for 1,190 (31.5%) participants, predominately G4 (Table 1). UACR data were available for 2,605 (69%) participants, with 920 (35.3%) exhibiting a UACR of ≥ 30 mg/mmol (Table S1 in S1 File).

There were 17 types of comorbidities recorded, with the number of comorbidities ranging from 0 to 12 per participant. A large proportion of participants (67.6%) fell into the moderate to very high comorbidity categories, with these groups distributed evenly across all CKD stages (Table 1). Of these 1,737 (68.1%) had an eGFR of ≥30 ml/min at referral (Table S2 in S1 File). Hypertension was the most common comorbidity, present in 77.5% of participants, followed by DM and cardiovascular disease (CVD), 44.9% and 40.7% respectively (Table 1).

Referral thresholds and outcomes

Around one third of participants (1,194, 31.5%) fulfilled the KHA eGFR threshold (<30ml/min) for specialist referral (Table S2 in S1 File). Among the 2,581 (68.4%) who did not fulfill the eGFR threshold, 216 (8.4%) progressed to KRT while 331 (12.8%) died without KRT (Table S2 in S1 File). Similarly, though 1,685 (64.7%) participants did not fulfil the UACR threshold of ≥30 mg/mmol for referral, of these 191 (11.3%) progressed to KRT while 221 (13.1%) died without KRT (Table S1 in S1 File).

Of the 2,605 participants with both eGFR and UACR data, 371 (14.2%) met both referral criteria, and 150 (40.4%) subsequently required KRT (Table S3 in S1 File). Conversely, 1,203 (46.2%) did not meet either threshold for referral, yet 79 (6.6%) of these went on to require KRT at some point during the period of follow up (Table S3 in S1 File).

Mortality and progression to ESKD

During the follow up period, 722 (19.1%) participants died, with the majority (598, 82.8%) dying prior to commencing KRT, while 775 (21%) participants progressed to ESKD, of whom 513 (66.2%) received KRT (Table 1). More than half of all deaths in the KRT group (63, 50.8%) occurred among participants from the low socioeconomic status group (Table S4 in S1 File). In the univariate Cox model, late referral was associated with a 27% lower risk of mortality (HR 0.73; 95% CI 0.43–1.26; p = 0.27). In the multivariable model adjusting for all variables except comorbidity score, the estimated mortality risk for late referral decreased to 14% (HR 0.86; 95% CI 0.49–1.49; p = 0.58) (Table S5 in S1 File). When the comorbidity score was added, the late referral estimate further decreased to 9% (HR 0.91; 95% CI 0.52–1.59; p = 0.74). The absence of an observed adverse effect of late referral on KRT mortality in our study is likely explained by confounding from comorbidity burden (HR 1.14; 95% CI 0.92–1.41; p = 0.27) (Table 2). The Kaplan–Meier curve indicated longer post-KRT survival in the late-referred group (Fig. S2 in S1 File); however, the log-rank test showed no statistically significant difference in survival between referral groups (p = 0.26). (Table S6 in S1 File). Advancing age was associated with increased mortality rate, with each 5-year increment corresponding to a 24% increase in the hazard of death (HR = 1.24; 95% CI 1.14–1.35; p < 0.01). Additionally, the presence of cardiovascular disease was associated with a significant increase in mortality rate (HR = 1.72; 95% CI 1.18–2.52; p = 0.01) (Table 2).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Univariate and multivariable Cox regression analyses (N = 513).

https://doi.org/10.1371/journal.pone.0338001.t002

Complementary competing-risks and sensitivity analysis

In the left-truncated competing risks analysis of the full cohort (n = 3,775), deaths occurring before KRT (n = 598; 15.8%) were included as competing events for post-KRT mortality (n = 124). After accounting for pre-KRT mortality as a competing risk in the fully adjusted, unrestricted Fine-Gray model, longer duration of pre-KRT care (time under care) showed a statistically significant but clinically minimal effect on post-KRT outcomes (SHR 1.0038 per additional month; 95% CI 1.0012–1.0064; p = 0.004). This corresponds to a 0.38% relative change in hazard per month, or approximately 4.7% over one year of care. However, mortality risk was primarily associated with baseline clinical factors. Each one-unit increase in comorbidity score was associated with more than a twofold increase in mortality risk (SHR 2.113; 95% CI 1.792–2.493; p < 0.001). Pre-existing cardiovascular disease (SHR 1.910; 95% CI 1.232–2.961; p = 0.004) and Indigenous status (SHR 1.931; 95% CI 1.193–3.127; p = 0.007) were also independently associated with increased risk. Age, sex, DM, hypertension, coronary artery disease, and social status were not independently associated (Table 3). To assess the impact of different referral thresholds, adjusted cumulative incidence functions were generated for 1, 3, 6, and 12 months of pre-KRT care (Figure S3 in S1 File). The resulting curves were closely aligned, indicating minimal differences in risk across these thresholds after accounting for competing risks and comorbidities.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Final multivariable fine-gray competing-risks regression for post-KRT mortality (N = 3,775).

https://doi.org/10.1371/journal.pone.0338001.t003

Referral timing and comorbidity

Of the 513 participants who commenced KRT, 60 (11.7%) were classified as late referrals. Among these, 32 (53.3%) were from low socioeconomic group, compared to 20 (33.3%) from high socioeconomic group (Table 4). Most participants who started KRT (397, 77.4%) (Table 4) and most who died (473, 65.5%) (Table S7 in S1 File) were in the moderate to very high comorbidity groups. Of those who died after starting KRT, (123, 99.2%) were from the moderate to the very high comorbidity groups, most of whom (108, 87.8%) had been referred early (Table S8 in S1 File). Only 15 (12.1%) of those who died post-KRT were late referrals, with 3 (20%) of these dying in the first 12 months of starting KRT (Table 4). The proportion of participants who progressed to KRT was similar across the moderate (15.2%), high (15.6%), and very high (16.1%) comorbidity groups, and all were higher than the proportion in the low comorbidity group (9.5%) (Fig. S4 in S1 File). Among those who progressed to KRT, the mortality rate increased progressively with higher comorbidity scores (Fig. S5 in S1 File). These findings suggest that comorbidity burden may be a stronger predictor of death than late referral alone.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Characteristics of participants who progressed to KRT.

https://doi.org/10.1371/journal.pone.0338001.t004

CKD stage and risk of progression

Participants in CKD stages 1–3 were found to be at greater risk of death before KRT than progress to KRT (55.4% vs 43.5%), whereas those in stages 4 & 5 were more likely to initiate KRT than die before KRT (56.5% vs 44.6%) (Table 1). Overall, mortality rates increased with advancing CKD stages (Table 1).

KFRE application and risk of KRT progression

Among the 1,743 participants who did not meet the KHA eGFR referral threshold and had available UACR data, descriptive cross-tabulation using 5-year KFRE risk thresholds classified 708 (40.6%) and 509 (29.2%) patients at the 3% and 5% risk levels, respectively. Within these groups, 144 (20.3%) and 126 (24.8%) progressed to KRT, while 91 (12.9%) and 65 (12.8%) died without receiving KRT (Table S9 in S1 File). Among participants who met the KHA eGFR threshold, KFRE classification identified a similar proportion of patients who progressed to KRT at both the 3% (99.6%) and 5% (96.6%) thresholds. The proportions progressing to KRT or dying were also similar across these thresholds (KRT: 31.0% and 31.9%; death: 15.6% and 14.9%; Table S10 in S1 File).

These descriptive findings suggest that the eGFR threshold and KFRE risk categories showed similar overall patterns in cohort distribution for KRT progression and pre-KRT mortality. However, KFRE additionally identified a group of higher-risk patients who were not captured by the eGFR threshold alone.

The case for a new definition of late referral: A risk-based approach

We propose that CKD referral timeliness should be prospectively redefined using risk stratification at the time of initial nephrology consultation. Current retrospective definitions, which classify referral timing according to variable intervals ranging from 1 to 12 months before KRT initiation, lack consensus, contribute to heterogeneity across studies, and offer limited value for clinical decision-making. In contrast, our findings suggest that baseline patient characteristics, particularly comorbidity burden and cardiovascular disease, are more strongly associated with post-KRT mortality than the conventional time-based measures of referral timing used in many previous studies.

Future prospective studies are needed to determine whether incorporating validated risk-prediction tools, such as the KFRE, at the time of referral can improve referral prioritisation and resource allocation while maintaining patient safety and equitable access to specialist care. Such an approach could enable clinicians to identify high-risk patients who may benefit from early vascular access planning and modality preparation, while allowing lower-risk patients to be managed with a greater emphasis on disease stabilisation, comorbidity management, and cardiovascular risk reduction.

Until such evidence is available, risk-based approaches should be considered a promising area for further investigation rather than a basis for changing current referral practices. Although these findings require confirmation in additional studies, they support consideration of a shift towards risk-based definitions of referral timeliness and CKD management.