Monitoring residual kidney function in haemodialysis patients using timed urine collections: validation of the use of estimated blood results to calculate GFR

Objective. With growing recognition of the benefits of preserving residual kidney function (RKF) and use of incremental treatment regimes, the incentive to measure residual clearance in haemodialysis patients is increasing. Interdialytic urine collections used to monitor RKF in research studies are considered impractical in routine care, partly due to the requirement for blood samples before and after the collection. Plasma solute levels can be estimated if patients are in ‘steady state’, where urea and creatinine concentrations increase at a constant rate between dialysis sessions and are reduced by a constant ratio at each session. Validation of the steady state assumption would allow development of simplified protocols for urine collections in HD patients. Approach. Equations were derived for estimating plasma urea and creatinine at the start or end of the interdialytic interval for patients in steady state. Data collected during the BISTRO study was used to assess the agreement between measured and estimated plasma levels and the effect of using estimated levels on the calculated glomerular filtration rate (GFR). Main results. The mean difference between GFR calculated with estimated plasma levels for the HD session after the collection and a full set of measured levels was 2.0% (95% limits of agreement −10.7% to +14.7%, N = 316). Where plasma levels for the session before the collection were estimated, the mean difference was 1.2% (limits of agreement −10.3% to +7.9%, N = 275). Significance. Using estimated levels for one session led to a clinically significant difference in the calculated GFR for less than 3% of the collections studied. This indicates that the steady state assumption can be used to estimate solute levels when determining GFR from timed urine collections. A pragmatic approach to monitoring RKF in HD would be for patients to collect for approximately 24 h before routine bloods are taken.


Introduction
Evidence of the benefits of preserved residual kidney function (RKF) in people on dialysis has been accruing for some years. The importance of RKF was first recognised in those treated with peritoneal dialysis, (PD), most notably in the Canada-USA study (Bargman et al 2001) where survival was 36% higher for each increment of 250 ml in urine volume, but subsequent studies have found similar benefits in the haemodialysis (HD) population. The Netherlands Cooperative Study of Adequate Dialysis (NECOSAD-2) observed that mortality was up to 17 times higher for anuric HD patients than those with measurable RKF, and more recently a US study found that a faster decline in RKF loss during the first year of HD was associated with increased risk of mortality (Termorshuizen 2004, Obi et al 2016. When routinely measured, the time for which clinically significant RKF can be maintained in HD patients has been shown to be comparable with those treated with PD (McKane et al 2002). Other advantages of preserved RKF include improved wellbeing, better quality of life (Merkus et al 1997) and the reduced need to remove high fluid volumes during dialysis sessions with the associated risks of intradialytic hypotension and cardiac stunning (Burton et al 2009).
Although HD guidelines do now allow dialysis dose to be adjusted if patients have sufficient renal clearance, it is still common practice for HD to be initiated at 'full-dose', typically 12 h/week in three sessions. An important barrier to a more individualised care in haemodialysis patients is that RKF is often not monitored in HD patients (Dasgupta et al 2016). While timed urine collections are carried out routinely in PD patients to measure RKF, they appear to be widely considered as impractical or inaccurate in HD patients, in part due to the cycling of plasma solute levels necessitating the taking of multiple blood samples. As well as impeding the implementation of incremental haemodialysis and the best use of resources, it is likely that this has hindered the conduct of research into interventions to help maintain RKF in HD patients.
Fluid depletion is known to accelerate loss of RKF, so interventions to improve fluid assessment could lead to better preservation of residual function. The BISTRO study (Bio-Impedance Spectroscopy to maintain Renal Output) was a randomized, controlled trial designed to determine whether using bioimpedance technology in setting the target post-dialysis weight can slow the loss of RKF in incident HD patients [Protocol: BMC Nephrol 2017] (Davies et al 2017). Although the primary outcome was time to anuria, of equal importance was the rate of decline of RKF, calculated as a glomerular filtration rate (GFR). The BISTRO protocol involved up to 15 GFR measurements per patient, each requiring a timed interdialytic urine collection and a set of blood samples from which to calculate the average plasma solute levels during the collection interval. This study provided the opportunity to evaluate the use of estimated plasma levels to calculate GFR.

Methods
Determination of residual GFR In dialysis patients, GFR is usually calculated from the mean of the renal urea and creatinine clearances (K r U and K r C), normalised to a body surface area (BSA) of 1.73 m 2 : The renal clearance for a solute from a timed urine collection is calculated using: where C urine is the concentration of the solute in the urine and C p_TAC is the time-averaged concentration of the solute in the plasma during the collection. All concentrations are in mmol l −1 . V urine and T urine are the volume and duration of the timed urine collection (in ml and minutes respectively). If a linear increase in solute concentration is assumed, C p_TAC is simply the concentration at the midpoint of the collection. For a full interdialytic urine collection, C p_TAC is the mean of the concentration of the solute at the end of session before the collection (C reb ) and the concentration at the start of the next session (C preNext ) Pre-and post-dialysis plasma levels (C pre and C post ) are required for the session before the collection to allow correction for rebound using the Tattersall formula (Tattersall et al 1996): where T d is the dialysis time in minutes and t p is the 'patient clearance time' for the solute (35 and 66 min for urea and creatinine respectively). Figure 1 shows the concentration profile during the interdialytic interval.
The standard protocol used to collect the data needed to calculate K r U and K r C in BISTRO was a full interdialytic collection which required patients to collect their urine throughout the interdialytic interval. A modified protocol was introduced to accommodate patients, for example those working, who wished to remain in the study but were not able to carry out the urine collections as specified above. For a 'partial interdialytic' collection, the three blood samples were taken as for the full interdialytic collection but the patient collected urine for a minimum of 24 h at the most convenient time during the interval between the sessions. They recorded the time they emptied their bladder before starting the collection and the last time they collected their urine in the canister. The equation used to calculate C p_TAC for a partial interdialytic collection is given in supplementary materials appendix A (available online at stacks.iop.org/PMEA/43/08NT01/mmedia).
Estimation of plasma solute levels using the steady state assumption When a patient is in 'steady state' with stable muscle mass and dialysis prescription, plus reasonably consistent daily protein and fluid intake, their plasma urea and creatinine concentrations will be reduced by the same fraction (R HD ) during each HD session and will increase with a constant gradient (G c ) between sessions (see figure 1). With a constant concentration gradient, plasma urea and creatinine throughout the week can be predicted using the pre-and post-dialysis levels from one session, as described in supplementary materials appendix B. This approach to predicting solute levels is adapted from the UK renal association guideline published in 1997 (treatment of adult patients with renal failure 1997). G c combines the solute generation rate and distribution volume, allowing the weekly concentration profile to be modelled without the use of sophisticated computation (Daugirdas et al 2009). If the steady state assumption is valid, allowing the use of estimated plasma levels, GFR can be calculated for a urine collection with blood samples taken at the session before or after.
Validation of the use of estimated compared to measured plasma levels in GFR calculations An audit of the procedure to monitor GFR during the BISTRO study provided the data required to assess the effect of using estimated plasma levels in two scenarios. For scenario A, the pre-and post-dialysis plasma levels were measured for the session before the collection and used to estimate the pre-dialysis levels for the next session. For scenario B, pre-dialysis plasma levels were measured for the session after the collection and used, together with pre-and post-dialysis levels from another session, to estimate the post rebound levels for the last session. For each patient with sufficient data, GFR was calculated for one collection with a complete set of urine data and measured plasma levels (GFR meas ), then recalculated using the estimated plasma levels for each scenario. The equations are for a patient on thrice weekly dialysis. The derivation of these equations, and the equivalent equations for twice weekly dialysis, can be found in supplementary materials appendix B.

Scenario A
For the scenario where pre-and post-dialysis plasma levels (C pre and C post ) were available for the session before the collection, the pre-dialysis levels for the next session were estimated using: where T int is the interdialytic interval during which the collection took place. R HD , the concentration reduction fraction during HD, is the ratio of the post-dialysis solute concentration corrected for rebound (using equation (4)) to C pre : For thrice weekly dialysis G c , the concentration gradient, is calculated from C pre and R HD using: where T intNext and T intLast are, respectively, the interdialytic intervals after and before the interval during which the collection took place. C p_TAC and the clearances for urea and creatinine were then calculated with the estimated C preNext and used to determine GFR estA .

Scenario B
For the scenario where the pre-dialysis plasma levels (C preNext ) were available for the session after the collection, the post-rebound levels for the last session were estimated using: As the BISTRO protocol did not include post-dialysis blood samples for the session after the collection, for scenario B, R HD was determined using pre-and post-dialysis solute concentrations for the session before another collection, ideally with the same treatment time (a 'matched' session).
In scenario B, G c is calculated from C preNextI and R HD from the matched session using: intNext HD 2 intLast HD int C p_TAC and the clearances for urea and creatinine were then calculated with the estimated C reb and used to determine GFR estB .

Selection of the validation dataset
The BISTRO study was registered in april 2016, (ISCCTN number: 11 342 007) and ethics approval obtained through the UK integrated research application system (project number 20 613) (Davies et al 2017). The inclusion criteria for BISTRO are described in detail in the published protocol. Deliberately pragmatic, there were few exclusions, but these included being unable to give consent, expected survival <6 months, certain multiple limb amputations and an inability to manage study procedures, including urine collections. Patients were screened within 3 months of commencing haemodialysis. Those who passed at least 500 ml of urine in the short interdialytic period (or had a calculated GFR >3 ml min −1 /1.73 m 2 , see below) were randomised and asked to carry out urine collections every month for the first 3 months, then every other month until they left the study.
Where available, one admissible interdialytic urine collection was included in the validation dataset for each patient who had been screened for the BISTRO study prior to the quality control audit in 2019. To be admissible for the validation exercise, the collection had to have a complete set of urine data and measured plasma levels. Collections were excluded where there was clear indication of incorrect labelling of pre or post blood samples, transcription errors or sample dilution. The BISTRO protocol did not require documentation of problems or changes in the delivery of dialysis that meant the patient would not be in steady state, but where such issues were recorded the collection was excluded.
For each patient with at least one admissible collection, the one closest to month 3 of the study was selected. Month 3 was chosen as it was hoped that problems with collections would have been identified and addressed by then, and that a wide range of GFR would be covered. The last admissible collection was selected for patients who had not yet reached month 3 of the trial, including those who were screened but did not meet the entry criteria. If the collection for month 3 was inadmissible or not fully documented, the closest admissible collection before or after month 3 was selected.
The matched sessions required for scenario B were taken from the closest available collection to the month selected for the validation dataset with plausible pre-and post-dialysis plasma levels for the session before the collection and a similar dialysis time. If there were equally close collections from before and after the selected month, the earlier one was used.

Statistical analysis
Descriptive statistics (mean and standard deviation) were used to compare the measured and estimated solute levels and the parameters derived from them (time averaged plasma concentration, renal clearance and GFR). The Bland and Altman method (Martin Bland and Altman 1986), where differences and 95% confidence limits of agreement are expressed as percentages of the values, was also used to assess the agreement and between GFR calculated using the full set of measurements and with the estimated levels. A difference of less than 10% or 0.5 ml min −1 /1.73 m 2 , whichever is larger, was considered to be clinically acceptable for monitoring GFR.

Results
Data from 346 patients recruited by 29 of the 32 BISTRO participating centres from across the UK (England, Wales, Northern Ireland and Scotland) were audited for data quality checks in 2019. The QC data included patients who did not have sufficient residual function to be randomised when screened, as well as patients who had been followed up for more than a year. 316 patients (median age 63, range 23-89 years; 71% male; 91% white) had at least one admissible interdialytic collection when their data was submitted. The admissible collection closest to month 3 of the study was selected for the validation dataset, as described in the methods. This was the month 3 collection for 142 patients. An earlier collection was selected for 77 patients who had not reached month 3 (including 20 who were not randomised). The closest collection to month 3 was selected for 97 patients who did not have an admissible collection at month 3 due to issues with the urine collection (25), lack of documentation or records showing a disrupted dialysis regime (15), missing blood results (40) and implausible plasma levels (17). The majority of the collections included in the validation dataset (93%) were full interdialytic collections and 95% of these were carried out over 2 d. The remaining 7% were partial interdialytic collections. The matched session required for the simulation in scenario B was available for 275 of the 316 patients.
For the selected collections, the mean GFR meas was 3.7 ml min −1 /1.73 m 2 (range 0.1-20.0 ml min −1 /1.73 m 2 ). Tables 1 and 2 show the mean and standard deviation for the solute levels, time averaged concentrations, renal clearances and GFR in the two scenarios. There was a tendency to underestimate the predialysis concentration for the session after the urine collection and to overestimate the post-rebound concentration for the session before. The discrepancy was halved in the time averaged concentration as would be expected when the estimate is combined with a measured value.
For scenario A, the mean difference between GFR estA and GFR meas as a percentage of GFR meas was 2.0% (95% limits of agreement −10.7% to +14.7%). For scenario B, the mean difference between GFR estB and GFR meas was −1.2% (limits of agreement −10.3% to +7.9%). Only 9 (3%) of GFR estA and 6 (2%) of GFR estB differed from GFR meas by more than 0.5 ml min −1 /1.73 m 2 (or 10% if this was larger). Figures 2 and 3 show the correlation and differences between the GFRs calculated with estimated and measured blood levels for scenarios A and B. A subsequent review of the outliers (difference >10%) showed changes in creatinine levels that suggested that these patients had missed a session or changed shifts and were not in steady state.

Discussion
Although there was no requirement to ensure that BISTRO participants were in a stable dialysis regime we found good agreement, over a wide range of RKF, between GFR calculated using the measured plasma levels and using 3.7±2.7 3.8±2.7 0.1±0.3 Table 2. Descriptive statistics (mean and standard deviation) for scenario B.

Scenario B (N=275)
Measured values only With estimated C reb Discrepancy C reb urea (mmol l −1 ) 6.8±2.8 7.1±2.9 0.3±1.4 C p_TAC urea (mmol L −1 ) 12.1±3.8 12.2±3.9 0.1±1.9 3.7±2.6 3.7±2.6 0.0±0.2 levels estimated by assuming the patients were in steady state. The difference between GFR meas and GFR calculated with estimated levels was clinically acceptable for 97% of patients when C preNext was estimated (scenario A) and 98% when C reb was estimated (scenario B). This validation exercise supports the use of protocols for urine collections where blood samples are taken at only one session. The logical choice would be to take blood samples at the session when the patient returns the canister having completed the collection. The agreement with GFR meas was slightly better for scenario B, even though GFR estB relied on data from a matched session at least a month earlier or later. Closer agreement might be expected if the HD reduction fraction had been determined using pre-and post-dialysis levels for the session after the collection, which would be the case if the collection was timed to end with the routine bloods.
Reducing the requirement for blood samples is an important step towards a pragmatic protocol for measuring RKF in HD patients. The other step is to reduce the collection time. The preference for full interdialytic collections for HD patients is based on concerns that the GFR is suppressed by haemodialysis, although there are few publications to support this. The most frequently cited paper indicating an increase in GFR during the interdialytic interval is from 1995 (van Olden et al 1995), in an era when target weight was routinely set to the lowest weight that the patient could tolerate and post-dialysis fluid depletion would have been the norm. To enable patients who could not collect for the full interdialytic interval to remain in the study, urine collections of at least 24 h were accepted for BISTRO. We were not able to make a comparison of partial and full interdialytic collections (this would require patients to switch to a new canister at a known time during the collection). However, collecting for a convenient period may be more accurate than collecting for the full Figure 2. Comparison between GFR est and GFR meas for scenario A (estimated pre-dialysis levels for the session following the urine collection). interval as the patient can empty their bladder naturally and they are less likely to forget to use the canister or overfill it. If acceptable to the patient, it is preferable to carry out the collection towards the end of the interdialytic interval as this minimises the time for which the urine is stored and, where blood samples are taken after the collection, the impact of any error in estimating the post rebound levels for the previous session. In patients with substantial RKF, collecting for the shorter interdialytic interval reduces the possible overestimation of GFR due to the non-linear (asymptotic) concentration rise (see supplementary materials appendix A).
In incremental dialysis, adjustments to the dialysis prescription are made using K r U, rather than GFR. The agreement between K r U, determined using equation (2), with the time averaged urea concentration calculated using measured and estimated plasma levels was similar to that for GFR. An alternative method for estimating C p_TAC for urea was published by Daugirdas (Daugirdas 2016) shortly after the BISTRO protocol was finalised. In this method, urine is collected for a minimum of 24 h ending just before the HD session where a pre-dialysis (and ideally post-dialysis) blood sample is taken. C p_TAC for urea is estimated using an empirical (best-fit) equation derived using formal urea kinetic modelling to construct the interdialytic plasma urea profile for a range of hypothetical dialysis regimes (Daugirdas et al 2009 where the urea reduction ratio, URR, is (C pre -C post )/C pre expressed as a percentage. For the full interdialytic collections used to validate scenario B, the Daugirdas equation gave a slightly higher C p_TAC (and a slightly lower K r U) than we obtained using the steady state assumption where URR was high (over 0.75).
The Daugirdas method offers a quick way to obtain K r U from a urine collection timed to coincide with the end of the interdialytic interval. For collections carried out earlier in the interval, or when GFR is required, the equations in this paper can be used. Our data suggest that urine volume should not be used as a surrogate for GFR or K r U. For the 60 patients with GFR meas between 3 and 4 ml min −1 /1.73 m 2 in the validation dataset (typically with K r U 2 to 3 ml min −1 ) the mean urine output was 680 ml d −1 but this varied from 220 to 1470 ml d −1 . The urine urea concentration in this cohort varied from 25 to 181 mmol l −1 . The factors contributing to this variation include hydration and fluid intake, which can vary for an individual as well as between patients, making residual urine output a poor indicator of RKF.

Conclusion
The good agreement between GFR meas and GFR calculated with estimated plasma levels for the session before or after the urine collection validates the steady state assumption, and the use of estimated levels when calculating GFR from timed interdialytic urine collections. This allows the protocol for urine collections to be simplified and, for the BISTRO study, justifies the use of estimates where blood samples have been missed.
To monitor RKF in HD patients, we suggest using timed collections of about 24 h in the dialysis interval before routine blood samples are scheduled. If this is not convenient, a pre-dialysis blood sample from the session after the collection can be used, together with pre-and post-dialysis plasma levels from a matched session. Other than noting the start and end times, the collection procedure is the same as for PD patients. Ideally, patients should keep to the same timing (which will usually be towards the end of the interdialytic interval) and not undertake collections if their dialysis regime has recently been changed or disrupted.
If a pragmatic protocol such as this is adopted in routine care, it will underpin awareness of the preservation of RKF and facilitate the implementation of incremental haemodialysis (Lindley and Tattersall 2022) For research studies looking at changes in GFR, such as BISTRO, this type of protocol will make the organisation of collections considerably easier.