Acute kidney injury (AKI) is a syndrome characterized by a rapid (hours to days) deterioration of kidney function.^[1,2]The incidence of AKI among critically ill patients and ST-segment elevation myocardial infarction (STEMI) patients is approximately 34% and 10%, respectively.^[3] AKI is an independent risk factor for the prognosis of critical illness and causes high mortality of 62%.^[4]Meanwhile, an epidemiological study in China indicated that approximately 1.4-2.9 million AKI patients were annually admitted; out of these patients, 28.5% were admitted to ICUs, and about 11.8% required renal replacement therapy.^[5] Despite being expensive, continuous renal replacement therapy (CRRT) is currently the most prevalent renal replacement therapy used in the ICU.^[6] A survey conducted in Jiangsu Province revealed that the average medical cost of CRRT for one AKI patient was 19,525 yuan.^[7] The cost of AKI treatment poses a significant burden to society and families. Thus, it is important to balance the treatment efficacy and its cost.

In the current clinical practice, CRRT filters are regularly replaced based on the manufacturer's instructions. Meanwhile, an unscheduled replacement can happen for various reasons, such as instances of filter clotting and catheter malfunction.^[8] Several studies indicate that the delivered dose of CRRT gradually decreases as the permeability of the filter membrane decreases with the extension of treatment time.^[9⇓-11] Nonetheless, whether filter replacement based on schedule affects patient outcomes remains to be further investigated. Besides, it is essential to identify appropriate indicators for filter replacement.

Therefore, in this single-centered, randomized, and crossover trial, we analyzed dynamic changes of serum levels of solutes and relevant indicators of the two most common CRRT modalities, i.e., continuous veno-venous hemofiltration (CVVH) and continuous veno-venous hemodiafiltration (CVVHDF), in order to identify an optimal indicator for changing the filter of CRRT thus improving treatment efficiency.

AKI patients requiring CRRT treatment in an ICU in the Second Affiliated Hospital of Soochow University between July 2017 and December 2018 were recruited.

Inclusion criteria included: (1) nonsurgical patients, age ≥18 years old; (2) AKI diagnosed based on the KDIGO clinical practice guidelines for acute kidney injury, 2012;^[12] (3) AKI patients with indicators that initiated CRRT, i.e., (a) stage 2 of AKI (supplementary Tables 1 and 2); (b) AKI patients combined with acute hypervolemic heart failure, acute pulmonary edema, hemodynamic instability due to septic shock,^[13] severe acid-base and electrolyte disorders, multiple organ dysfunction that necessitated CRRT.

Exclusion criteria included: (1) patients with chronic renal failure or end-stage renal disease receiving maintenance dialysis; (2) the required treatment parameters could not be used because of the disease condition; (3) during the treatment process, the treatment mode and parameters needed to be changed according to the disease condition; (4) during the treatment process, the patient condition significantly changed, hence unsuitable to continue in the clinical trial; (5) the crossover remained unfinished because of the withdrawal or death.

Using a crossover design,^[14] the recruited AKI patients were randomly divided into groups A and B using random numbers and envelope methods. Patients in the group A firstly received CVVH followed by CVVHDF. Conversely, patients in group B firstly received CVVHDF followed by CVVH. Since CRRT treatment did not have an apparent carry-over effect, non-treatment time for 30-40 minutes was required to switch from CVVH to CVVHDF and vice versa, which could be a sufficient wash-out period of the previous treatment.^[15]

Femoral vein catheterization was performed in all patients. The prescribed dose of both CVVH and CVVHDF was 35 mL/(kg∙h) (post-dilution). The rate of dialysate flow and replacement fluid flow of CVVHDF was 1:1. The filter and circuit were pre-washed using 3 L normal saline with 37,500 U heparin sodium, maintained at 3-5 U/(kg∙h) in the process of CRRT. The activated partial thromboplastin time (APTT) of the filtered blood was maintained at 1.5-2.0 times of normal level by adjusting the dose of heparin. Filter replacement indicators were: transmembrane pressure (P_TM) increase (≥300 mmHg [1 mmHg=0.133 kPa]), and pre-filter pressure (P_PRE) increase (≥300 mmHg), filter and circuit use time reaching the specified length (≥60 h).

Observational indices included: (1) clinical characteristics: Acute Physiology and Chronic Health Evaluation II (APACHE II) score, Sequential Organ Failure Assessment (SOFA) score, serum level of urea nitrogen, creatinine, β₂-microglobulin, and cystatin C before treatment; (2) solute removal efficiency: delivered dose (specific serum solute delivered clearance calculated using it's measured blood and effluent concentration), serum solute concentration, P_PRE and P_TM at different time points; (3) patient outcomes: duration of mechanical ventilation, length of ICU stay, renal function recovery rate, and the in-hospital mortality rate.

Blood samples were collected every 12 h after CRRT treatment initiation (the first sample was delayed 30 min to avoid pre-wash fluid contamination). P_PRE and P_TM at the time of collecting blood samples were recorded. Urea nitrogen, creatinine, β₂-microglobulin, and cystatin C in blood and effluent samples were detected for calculating the delivered doses. The detailed equipment and materials were listed in the supplementary Table 3.

Formulas of calculating prescribed dose^[11,16] and delivered dose^[17] were as follows:

(1) Prescribed dose (K_P, mL/[kg∙h])

(2) Delivered dose (K_D, mL/[kg∙h])

The above formula was used for the calculation of the delivered dose of urea nitrogen. Solutes with other molecular weights were similarly calculated. In this formula, Q_D was the dialysate flow rate (mL/h), Q_UF was the ultrafiltration rate (mL/h), W was the body weight of the patient (kg), Q_R was the replacement liquid flow rate (mL/h), Q_NET was the net ultrafiltration rate (mL/h), UN_F was the urea nitrogen level in the ultrafiltration, UN_B was the urea nitrogen level in the blood (mmol/L), and Q_EFF was the effluent flow rate (mL/h).

The SPSS 23.0 software was used for statistical analyses. Enumeration data were expressed as number (%) and compared using the χ² test. Fisher's exact test was performed in the condition where total frequency <40 or the frequency <1. Normally distributed measurement data were expressed as mean±standard deviation and compared using the single factor t-test; otherwise, they were expressed as median (interquartile range), and independent and paired samples were compared using Mann-Whitney U-test and Wilcoxon signed-rank test, respectively. Partial correlation analysis was performed to identify a correlation between delivered dose and P_PRE and P_TM. Receiver operating characteristic (ROC) curves were generated to assess the sensitivity and specificity of P_TM in predicting the time to change the filter. A P-value <0.05 was considered statistically significant.

A total of 90 critically ill patients were recruited, whereas 26 were excluded. In total, 64 cases were randomly divided into groups A and B. During the treatment, 12 cases gave up the treatment, and 2 cases altered treatment parameters. Consequently, clinical data of 14 cases were not analyzed. Eventually, 50 eligible cases were analyzed, 27 cases in the group A and 23 in the group B. Flow chart showing the recruitment and research method is illustrated in the supplementary Figure 1. No significant differences were noted in sex, age, APACHE II score, SOFA score, serum urea nitrogen, creatinine, β₂-microglobulin, and cystatin C between groups (all P>0.05). There were also no significant differences in patients' duration of mechanical ventilation, length of ICU stay, renal function recovery rate, and in-hospital mortality rate between groups (all P>0.05) (Table 1).

With the prolongation of CRRT, the delivered doses of small- and medium-molecular-weight solutes were decreased when treatment ceased (filter and circuit replacement), compared with delivered doses at the initiation of CRRT treatment, particularly those of medium-molecular-weight solutes (supplementary Figure 2).

In CVVH mode, delivered doses of urea nitrogen, creatinine, β₂-microglobulin, and cystatin C decreased by 2.3% (0.4%-4.0%), 1.9% (0.2%-4.2%), 44.1% (15.8%-62.6%), and 45.5% (22.5%-70.3%), respectively (all P<0.05).

In CVVHDF mode, the delivered doses of urea nitrogen, creatinine, β₂-microglobulin, and cystatin C decreased by 3.5% (0.5%-5.1%), 2.5% (0.5%-6.2%), 41.5% (16.9%-54.8%), and 45.1% (1.8%-68.4%), respectively (all P<0.05).

Serum levels of small-molecular-weight solutes gradually decreased during the whole process of CRRT, whilst those of medium-molecular-weight solutes sharply decreased in the early stage of the treatment, then the magnitude of the decrease smoothly reduced in the middle and late stages. Moreover, the rebound of serum medium-molecular-weight solutes was observed in most patients in the late stage of CRRT treatment, which was more frequently observed with the prolongation of CRRT in both CVVH and CVVHDF mode. However, no significant difference between the two modalities was detected (P>0.05) (Table 2).

No significant correlation was noted between P_PRE and delivered dose of either small-molecular-weight solutes or medium-molecular-weight solutes (both P>0.05). P_TM was significantly correlated with the delivered dose of medium-molecular-weight solutes, rather than that of small-molecular-weight solutes (P<0.05). In CVVH mode, P_TM was negatively correlated with the delivered dose of β₂-microglobulin (r= -0.458, P<0.01) and cystatin C (r= -0.226, P<0.01). In CVVHDF mode, P_TM was negatively correlated with the delivered dose of β₂-microglobulin (r= -0.503, P<0.01) and cystatin C (r= -0.296, P<0.01) (supplementary Figure 3).

The time point when a rebound of serum solutes occurred was considered the optimal time to change the filter of CRRT. As stated above, P_TM was negatively correlated with delivered doses. To further investigate the predictive potential of P_TM, ROC curves were generated to assess the sensitivity and specificity of P_TM in predicting the rebound of β₂-microglobulin and cystatin C.

The area under the curve (AUC) of P_TM in predicting the rebound of β₂-microglobulin was 0.651 (P<0.01), and the threshold of P_TM was 146.5 mmHg (sensitivity=0.479, specificity=0.811) (Figure 1A).

The AUC of P_TM in predicting the rebound of cystatin C was 0.717 (P<0.01), while the threshold of P_TM was 146.5 mmHg (sensitivity=0.512, specificity=0.871) (Figure 1B).

In this study, we found that delivered doses of solutes with different molecular weight decreased in both modalities of CRRT when treatment ceased (filter and circuit replacement), compared with delivered doses at the initiation of CRRT treatment. Nonetheless, during all the CRRT courses, we did not detect a rebound of serum concentration of small-molecular-weight solutes. The delivered doses of medium-molecular-weight solutes decreased far more significantly than those of small-molecular-weight solutes; as a consequence, a rebound in their serum concentration was noted in most patients.

During CRRT treatment, clinicians and nurses often maximize the filter life span to minimize the treatment cost. Nevertheless, prolonged filter use causes membrane rupture, hence harming the patient. Thus, filters are often regularly replaced in the absence of clotting as per the experience or manufacturer's safe use time limits. A recent study reported that the use of citrate anticoagulation and non-convective modalities in reducing hemoconcentration prolonged the use of filters, thereby minimizing the treatment cost.^[18] Nonetheless, our study and several others^[9⇓-11] indicate that the delivered dose of CRRT gradually decreases as the filter membrane permeability decreases with the extension of treatment time. As such, the filter should be replaced before the delivered dose decreases. However, how often the filter should be replaced remains unresolved. Regarding the economics of treatment, a few studies minimized the cost of treatment by reducing the amount of replacement fluid^[19] or using cheaper replacement fluid.^[18] Here, we consider that instead of sacrificing the treatment dose of patients or even treatment safety to reduce costs, it is sustainably economical to flexibly use filters according to the treatment effect on patients. Thus, based on our findings, for patients requiring the removal of small-molecular-weight solutes, the filter should be used as long as possible within the manufacturer's safety limits. This lowers the treatment costs and increases treatment efficiency since the serum concentration of small-molecule-weight solutes continuously decreases during the treatment course. On the other hand, for patients requiring removal of medium-molecular-weight solutes, the filter should be replaced when the efficacy of the filter decreases to the point where it is unable to regulate the serum concentration of medium-molecular-weight solutes, and it's the point where patients' serum medium-molecular-weight solutes concentration rebounded. This occurs in the second half of the treatment; at this point, the treatment should ideally be stopped because the serum solute concentration of the patients cannot be effectively regulated due to filter function impairment.

Our study confirmed that P_TM was significantly and negatively correlated with the delivered dose of medium-molecular-weight solutes in different treatment modalities; i.e., an increase in P_TM indicated a decrease in the delivered dose. ROC curves demonstrated that P_TM could precisely predict the rebound of serum level of cystatin C. Notably, the production rate of serum cystatin C is relatively constant, while that of β₂-microglobulin is affected by multiple factors.^[20] Thus, cystatin C is a promising indicator for serum levels of solutes. When P_TM reaches 146.5 mmHg, the delivered dose of medium-molecular-weight solutes may be insufficient to regulate serum levels of solutes. Therefore, the time at which P_TM reaches 146.5 mmHg may be the optimum time to change the filter among patients requiring the removal of medium-molecular-weight solutes.

Noteworthily, an important feature of ICU patients is their heterogeneity. For instance, the primary diseases were not identical. Besides, the treatment for their primary diseases was inconsistent or individualized even with the adoption of similar CRRT treatment parameters and modalities. In addition, pathophysiologic conditions including renal functional status are constantly changing in different patients, and the response to treatment varies in different stages of the disease. These potentially trigger variability in test results, specifically in small sample sizes.^[21] A crossover experimental design was used to markedly eliminate the effect of differences in the underlying characteristics of different patients on the experimental results.^[14]

This study has some limitations. First, it is a single-center study with a relatively small sample size (n=50). Although a crossover design was adopted, there was a possibility of errors due to the small sample size. In addition, numerous treatment and monitoring parameters of CRRT have been reported, and whether additional reliable indicators of filter replacement are available remains to be investigated.

In summary, the filter can be used during CRRT as long as possible within the manufacturer's safe use time limits to therapeutically remove small-molecular-weight solutes. Meanwhile, P_TM of 146.5 mmHg may be an optimal indicator for changing the filter when removing medium-molecular-weight solutes during CRRT.

Funding: The study was supported by Kunshan Science and Technology Special Fund (Social Development Category, KS18040).

Ethical approval: This trial protocol was reviewed and approved by the ethics committee of the Second Affiliated Hospital of Soochow University.

Conflicts of interests: The authors declare that there is no conflict of interest.

Contributors: CH designed the research and wrote the paper. All authors contributed to the design and interpretation of the study and to further drafts.

All the supplementary files in this paper are available at http://wjem.com.cn.