The literature on whether warnings and interruptive alerts in EHRs positively impacts patient care is mixed. A small, non-significant effect on RBC transfusion has been demonstrated in a previous study [16], while other studies revealed a significant positive effect, especially on RBCs [17,18,19,20,21,22], while others reported a non-significant decrease in inappropriate fresh frozen plasma and platelets transfusion [21, 23]. Goodnough et al. [24] reported an improvement in many outcomes after introducing blood component-related CDS rules. These include a decline in RBC utilisation, a decrease in mortality, length of hospital stay, 30-days readmission rate, and in the total hospital transfusion-related costs of RBCs. A randomised control trial of a CDS intervention has demonstrated effectiveness of CDS involving all blood components [25].
In line with national and international guidelines as supported by evidence from clinical trials, the haematologists at our institution requested changes to the EHR to educate users and attempt to reduce inappropriate transfusions, as implied by these guidelines. Guidelines of course cannot infer that all transfusions out with them are inappropriate but it is likely that a large proportion would be. Initial levels for inappropriate transfusions were lowered over time reflecting national guidance [6]. We were also very interested in studying the large groups of patients in CC and HO as distinct groups for two reasons, firstly these patients receive a significant proportion of RBC and platelet transfusions in the trust (74.23% of RBC transfusion and 90.27% of platelets transfusion) and secondly, prescribing behaviour is likely to be different in these groups for reasons discussed below. Of note 8% of platelet transfusions were given to patients outside of critical care or haematology/oncology. Although many patients will be specialty specific, some patients with chronic haematological conditions or cancers will be cared for by other specialities and therefore be spread across the hospital.
There are no studies that help answer the question about why different specialties react differently to warnings, and this is perhaps one of the most interesting outcomes. One could imagine that haematologists are unlikely to think that they need automated prompts to inform their transfusion decisions, and junior doctors in these areas may defer to senior colleagues. In ICU more exceptions are made to usual practice in our experience. This is a relevant area of research as ‘one size fits all’ for CDS may need to be nuanced across hospital specialties. We continuously update and modify the EHR based on new evidence, user feedback and behaviours.
There are important reasons why RBC and platelet transfusions may be appropriate above the warning thresholds set in the EHR. Platelets may be needed in active bleeding and prior to invasive procedures. The Trust is a major trauma centre, frequently treating patients with major haemorrhage.
We found that the warnings did change user behaviour. The effect was most obvious outside the specialist settings of CC and HO, where there was a significant reduction in RBC transfusions given to patients with haemoglobin of over 70 g/L. While it has been longstanding practice to transfuse RBC at somewhat arbitrary thresholds, evidence from large randomised controlled trials over the last 20 years supports use of restrictive thresholds [26].
Blood transfusion can be utilised differently in HO where many patients will have chronic, transfusion dependent anaemia and be transfused to a personalised threshold depending on their symptoms. In oncology, patients are frequently transfused to a haemoglobin > 100 g/L prior to starting chemotherapy, although this practice is not evidence based and contravenes local and national guidance. Due to the lack of evidence to support use of restrictive thresholds for RBC transfusion in patients undergoing intensive treatment for haematological malignancy, trust guidelines recommended a threshold of 80 g/L in this group until 2019.
In CC we did not see a significant effect on transfusion after introducing warnings. In fact the feedback from clinicians in this area was negative in respect to these warnings, and we were repeatedly asked to remove them. Utilising questionnaires and focus groups users stated that this warning contributed to alert fatigue, and was perceived as non-contributory to patient care. In some situations, such as acute coronary syndrome, guidelines exclude the 70 g/L threshold, which may influence prescribing in cardiac CC [6]. Furthermore, there are randomised controlled trials in patients undergoing cardiac surgery which largely support the safety of restrictive thresholds but the outcomes of which are subject to debate [27, 28].
This contributes importantly to understanding how different areas will react to CDS dependent on relevance to the patient group, and familiarity of the clinicians with the issue they are faced with. Clinicians who often prescribe blood are seemingly less likely to take advice or change behaviours in response to computerised warnings.
Of note the majority of blood transfusions occur in HO and CC. Introduction of warnings has an effect on user prescribing in these areas, but this is not sustained. This may be linked to alert fatigue—outside CC and haematology where users do not encounter warnings very often, it is possible that they take more notice of them [29]. Considering that there is evidence on restricting transfusion in ICU, this is a relevant target for future studies to understand how CDS can create sustained behaviour change in this environment.
In CC overall the proportion of platelet transfusions given with a pre-transfusion platelet count > 10 × 109/L fell during the study time but paradoxically the warnings appeared to transiently increase transfusions. In HO, platelet transfusions significantly fell in association with introduction of warnings. Since this is a largest area of use, clinician behaviour change here is important.
EHR can positively influence clinician behaviour but must be designed by the clinicians themselves and effects must be studied over time, as initial positive effects may wane. Where warnings are frequent, alert fatigue is likely and will not only result in warnings being ignored, but will frustrate and slow down clinicians. The EHR in our institution is designed in conjunction with the clinicians themselves and developed in an iterative manner. The software is built and controlled by the organisation so where clinicians do not approve of, or change their mind about, EHR design, the software is modified. Studies of this type check over time that behaviour changes are sustained. All EHR changes are accompanied by education and communication to users. The software build is safeguarded by clinicians involved in its management who have previously published on the effects of alert fatigue. This is important when institutions are considering CDS within EHR.
Limitations
There are several limitations to this study. Importantly, the study includes all patients other than those in ED and those on a renal dialysis programme. Patients excluded from the 70 g/L threshold in national and local guidance, such as those who are chronically transfused, those with acute coronary syndrome, and those with major haemorrhage have been included in the data. Similarly patients undergoing procedures or who have active bleeding may receive appropriate platelet transfusions at thresholds higher than those given. No individual patient data were reviewed to examine whether transfusions were given at higher thresholds for valid clinical reasons. This would explain the seemingly very high proportion of patients who received a transfusion above accepted thresholds in our study. It is reasonable to assume that the proportion of patients receiving transfusions for those reasons would not change over time.
Although we have used a methodology that allowed us to analyse the longitudinal effect of the interventions [14], other confounding factors may have influenced RBC and platelet prescribing behaviour over the course of the study. Importantly, the introduction of Patient Blood Management, with the focus on evidence based practice and subsequent adoption of more restrictive transfusion thresholds worldwide and subsequent downward trends in blood use, may have had an impact on the prescribing practices of doctors in our institution, irrespective of alerts on the PICS [30]. In 2015 a dedicated transfusion consultant was appointed at our trust for the first time, and subsequently there has been more intensive clinical support for transfusion training and delivery.