Impact of a computerized decision support tool deployed in two intensive care units on acute kidney injury progression and guideline compliance: a prospective observational study

The study was conducted in two ICUs at University Hospitals Bristol, UK, a closed format tertiary medical and surgical ICU (GICU) and closed format tertiary cardiac ICU (CICU). GICU admits over 1200 patients per year and CICU is a regional cardiac surgical center with over 1800 admissions a year. Both units have been using the Philips IntelliSpace Critical Care and Anesthesia (ICCA) system since 2015. This system automatically collects all information relating to patient care including laboratory data and hourly vitals and observations. It also allows the development and deployment of clinical decision support software and analysis tools.

The study was divided into two phases—the control phase when no AKI guidelines and care bundle was shown to clinicians and the intervention phase when AKI guidelines and care bundle was available to clinicians via ICCA. The control phase consisted of retrospective data from consecutive adult patients admitted to GICU and CICU over 18 months for more than 24 h. The intervention phase consisted of adult admissions to the same units over 12 months for more than 24 h. For both phases, encounters were excluded from analysis if (1) age, race or gender information was not available, (2) age at admission > 90 years, (3) had acute kidney injury on admission defined as oligoanuria in the first 6 h of admission, (4) patients received renal replacement therapy (RRT) during ICU stay. (Patients receiving RRT were excluded from the analysis of AKI scores due to confounding of the urine output and creatinine values by the RRT, giving a false score output from the algorithm. The use of RRT was displayed as AKI Stage 3 in the decision support tool). Additionally, encounters where insufficient data were available to calculate AKI score (urine output, body weight, serum creatinine) were excluded. The institutional review authority deemed this investigation a quality improvement project and waived the requirement for patient consent. Figure 1 shows the patient selection process for the control and intervention phase.

The intervention consisted of three components—(1) automated AKI guidelines, (2) AKI care bundle tailored to workflow and (3) education and training about AKI guidelines and the care bundle.

The automated electronic AKI staging guidelines were based on AKIN [13]. The automated electronic AKI staging algorithm has been previously developed and validated [12]. The AKI stage is determined to be the maximum of the most recent values of AKI using urine output staging (AKI UO) and AKI using serum creatinine staging (AKI Cr). AKI UO is calculated as described in the AKIN guidelines using weight normalized hourly urine output calculated over different time periods for different stages. AKI Cr is also calculated per guidelines, based on the increase in measured serum creatinine from baseline serum creatinine. This algorithm was integrated into ICCA (version H.02.01). AKI Cr and AKI UO each were calculated every time; a new measurement of serum creatinine and a new measurement of urine output, respectively, were charted in ICCA.

The AKI care bundle was implemented in ICCA and comprised of five following elements;

These five elements are mapped to elements of the workflow in Fig. 2. The first four elements of the care bundle were implemented in both units. The AKI flow sheet and advisories are visible to all members of the multidisciplinary team (MDT) (doctors, nurses and pharmacists). Advisories can be acknowledged but no aspect of the decision support tool is mandatory. The AKI bundle form and AKI order set are available to doctors. In GICU, the dashboard is visible to all members of the MDT. The dashboard was not implemented in CICU due to lack of resources. In addition, the admission document for both units was modified to include charting of baseline serum creatinine and listing nephrotoxic medications currently prescribed according to a predefined list.

The third component of the intervention was training and education for the clinicians in the two units. To facilitate this, a video was created to explain how AKI scores are calculated and how to use the care bundle. This video was shared with the clinicians and made accessible in both units. There was no additional teaching on AKI during the study but rotating junior doctors in GICU all received instruction on AKI management as part of their scheduled teaching program. Nursing staff had no additional education on AKI management.

There were no other changes in clinical practice during the study, DVT prophylaxis protocols were the same, and there were no additional staff including specialist AKI nurses or additional pharmacists. The use of the new workflow was highlighted as part of daily safety briefs during morning rounds.

Data were extracted from CCIS to compute the AKI score and evaluate the outcomes. All data were anonymized to remove identifiable information and all timestamps were shifted. Extracted data included demographic information such as age, gender, ethnicity, ICU admission and discharge time; serum creatinine measurements, baseline serum creatinine at admission and urine output for calculating the AKI stage. Enoxaparin doses and estimated glomerular filtration rate (eGFR) for evaluating enoxaparin outcomes were also extracted.

Continuous AKI scores were calculated for the control phase using electronic sniffer algorithm and have been previously described in a previous publication [14]. AKI scores were calculated anytime a new urine output or a new serum creatinine measurement was charted. To calculate AKI Cr, a baseline serum creatinine value is needed. In the control phase, this value was mostly not available; therefore, the MDRD equation [15] was used to estimate baseline creatinine for the entire cohort. In the intervention phase, majority of the patients had baseline serum creatinine charted. For the rest of the patients, MDRD equation was used to calculate baseline creatinine values. To calculate AKI UO, urine output values were normalized by the most recently measured weight (daily or admission weight).

The primary outcome of the study was the proportion AKI patients developing a worse stage of AKI during their stay. Additionally, the proportion of correctly administered enoxaparin doses was used as a measure of clinician awareness of AKI & guideline compliance. The secondary outcomes are shown in Table 1. The time-series values of AKI (AKI UO and AKI Cr), eGFR and administered enoxaparin dose were used to compute these outcomes. The outcomes were then compared between the control phase and the intervention phase. The analysis further done for each ICU unit separately and results are reported.

To determine the primary outcome, i.e., increasing AKI risk from Stage 1, the number of encounters where there was at least one increase from Stage 1 to a higher AKI stage, was calculated. In addition, the number of encounters where there was at least one increase from Stage 2 to Stage 3, was also calculated.

To evaluate guideline compliance, the correct dose was defined as an amount equal to or less than the recommended daily enoxaparin dose based on renal function (See Additional File 1: Table S1). To evaluate if the correct dose was given, we extracted the dose and timing information for all nonzero doses of enoxaparin. The closest preceding eGFR value was also extracted. Every enoxaparin dose was classified as correct or incorrect based on the preceding eGFR value. The total number of correct and incorrect doses per encounter and for each phase of the study was calculated. Any enoxaparin dose given before eGFR was available was removed from the analysis. Patients who received dialysis or weighed more than 100 kg were also excluded from this analysis. In addition to evaluating by dose, we also computed the number of admissions in which correct dose of enoxaparin was given for the entire ICU stay.

The primary outcome was evaluated after control phase data were extracted and have been published previously [14]. The primary outcome-rate of AKI stage increase was used to power the study. The proportion AKI patients progressing to a worse stage of AKI was used to determine the study sample size. To show a reduction in 15% of the number of patients progressing to a worse AKI stage, a study size of 6000 admissions (3000 controls and 3000 main phase) is needed for power of 0.86 and α error of 0.04 (using Fisher’s exact one-tailed test). Based on available information on the number of patients per year, it was estimated that 3000 patients could be recruited in 12 months. This includes a portion of cohort which might be excluded from analysis due to insufficient data or not meeting inclusion criteria.

Kaplan–Meier survival curves were generated using time to AKI as the event of interest. Patients who were discharged or developed AKI each day were removed from analysis to generate these curves. This analysis was done for both phases.

In control phase, out of 5140 consecutive ICU admissions over 18 months, 2523 ICU admissions were selected for data analysis and comparison. In the intervention phase, 2932 consecutive ICU admissions were enrolled into the study. After applying the exclusion criteria, 2521 ICU admissions were used for analysis (Fig. 1). Table 2 shows the patient cohort demographics in the two phases of the study. The two cohorts were similar in age, ICU length of stay, gender distribution and distribution in the two units. Admission weight was significantly higher in the intervention phase and in the intervention phase the proportion of black patients was significantly higher, although a very small proportion of the population. The proportion of patients getting RRT in ICU was significantly higher in the intervention phase (7.9% in intervention phase vs 6.5% in control phase), the proportion of readmissions was significantly higher in the intervention phase.

Entering baseline serum creatinine value was a part of the intervention. The proportion of admissions with a value increased from 1.6 to 58% in the intervention phase.

In the addition to these demographic features, we also compared demographic, admission and discharge features in each unit (Table 3). Hospital length of stay was significantly longer in the intervention phase in both units. ICU length of stay was similar in the GICU in both phases. In CICU, the ICU length of stay was longer by 0.4 days in intervention phase, which was statistically significant. The proportion of medical and surgical admissions was similar in the two phases in GICU; however, the nature of surgery was statistically different. In GICU, the proportion of emergency/urgent surgery was significantly higher, while the proportion of urgent and elective/scheduled surgery was significantly lower in the intervention phase. In CICU, the proportion of urgent surgeries was significantly lower, while the proportion of elective/scheduled surgeries was significantly higher in intervention phase. Among outcomes, mortality was significantly lower and readmissions significantly higher in CICU. No significant difference was observed in the GICU.

The proportion of Stage 1 AKI patients progressing to higher AKI stages, decreased in the intervention phase (42–33.5%, p = 0.002) (Fig. 3). In the GICU, the proportion of deteriorations decreased from 50.4% in control phase to 32% in intervention phase (p < 0.001), while in CICU the proportions decreased from 36.8% in control to 34.4% in intervention (p = 0.5). The proportion of encounters in Stage 2 increasing risk to Stage 3, decreased from 21.3% in control phase to 11.8% in intervention phase (p = 0.005).

The adherence to eGFR-based enoxaparin dosing guidelines was evaluated as described above and see Additional file 1: Table S1. The proportion of incorrect doses decreased from 1.72% in control phase to 0.6% in the intervention phase (p < 0.001). This decrease was largely from the CICU where the proportion of incorrect doses decreased from 2.62% in control phase to 0.72% in the intervention phase. The GICU, in contrast had lower proportion of incorrect doses, which decreased from 0.68 to 0.43% (Fig. 4).

The analysis of wrong doses grouped by eGFR value and dose revealed that the majority of the incorrect dosing occurred in eGFR range 20–30 ml/min/1.73 m² when 40 mg dose of enoxaparin was given (correct dose is 20 mg). The second frequent wrong dosing occurred in the eGFR range < 20 ml/min/1.73 m² when 40 mg of enoxaparin was given (correct dose is 0). For both these cases and other cases, the number of wrong doses decreased in both units in intervention phase.

The Kaplan–Meier curves plotting the cumulative proportion of patients with no AKI over days in ICU are shown in Fig. 5. These curves are shown for the first 10 days of ICU stay for the control and intervention phase. The cumulative proportion curve for the intervention phase is higher than the control phase indicating that in each successive day fewer patients developed AKI in the intervention phase. The table below shows the number of patients with no AKI at different ICU stay days.