Our survey evaluated answers given by nephrologists regarding their awareness of existing CDSS, their openness to use CDSS, their knowledge of how CDSS can assist them in everyday patient care, and an assessment of prediction algorithms as well as literature support being part of a future nephrologists’ CDSS. The work-up of this survey will help in addressing possible barriers for implementation of CDSS in a nephrologist’s medical routine and will therefore facilitate an optimal development of CDSS in the nephrology field. Past studies have shown that the identification of a specific target group is an important aspect of CDSS design. Notably, specialty physicians reported an increased need for a knowledge source that helps to answer complex context specific clinical questions, which implies that they are more likely to benefit from a specified CDSS [29]. Surveys are a useful and valuable tool in determining needs of researchers developing tools, CDSSs on the one hand and research users/physicians on the other hand [30].
In everyday clinical practice, treating nephrologists are confronted with a large amount of patient data as well as a multitude of research papers and guidelines. Moreover, patient care needs to be individualized as well as based on these newest guidelines and extensive research has been done on the facilitators and barriers of adherence to guidelines in patient care [31, 32]. In recent years, a growing number of mobile-health technology in the management of chronic diseases has been observed [33, 34]. These applications are mainly aimed at patients as end-users and primarily achieve pure data collection [35]. The integration of data accumulated through these apps in clinical nephrology practices is limited by: 1. A diversity of programs, 2. Data entries, that are optimized to the patient perspective and not the physician’s perspective, 3. A lack of compatibility with clinical data/data processing software (adequate infrastructure) [36]. In addition, highly sought after and need of prediction algorithms of possible adverse events and disease progression require models developed and validated in CKD patient cohorts. Moreover, a transparent presentation of outcomes from these prediction models is of high importance for the treating nephrologist. Taken together, these limitations have led to low ratings of CDSS use by nephrologists [37, 38]. Early integration of physicians into the development of a CDSS is therefore essential for later implementation in clinical practice.
Nephrologists’ experiences with CDSS in general
Little is known about the experiences of German nephrologists with CDSS. Our survey showed that 80% of all participants rarely or never use CDSSs, but about 80% of participants would find a personalized CDSS helpful for CKD patient care and over 70% of participants believe that it would improve patient care as well as their work efficiency. In the past, many examples of helpful CDSSs can be found supporting this notion [12, 39–42]. Taken together with the knowledge that specialty physicians in particular profit from CDSS support [29], the development of a specialized CDSS for nephrologists supporting them in personalized CKD patient care, seems overdue.
Nephrologists’ expectations towards a helpful CDSS
In our survey, most nephrologists generally agreed that the most important topics are accessibility to newest guidelines and other important secondary information essential for CKD patient care, but also management of patient data from previous visits. This is in line with data obtained from other groups working on facilitators for the implementation of CDSSs. Varonen et al. were able to show that physicians are mostly interested in managing the mass and complexity of information available when thinking of CDSSs [43]. Another important aspect of this module was the resistance to change of a nephrologist’s workday. Here, 40% of all participants were opposed to any changes, an aspect that has been seen during the implementation process of CDSS before [44], bringing up the necessity to carefully describe and collect contextual factors important for practice changes before CDSS implementation [45, 46]. Overall, nephrologists did recognize the usefulness of CDSS, which is a common theme on the perception of CDSS by health care professionals [47]. But long-term implementation of CDSS may pose a challenge that might be tackled by the implementation of behavior change interventions [48].
Evaluation of the importance/usefulness of prediction algorithms within CDSS
CKD patients constitute a heterogeneous group, making the development of useful prediction models challenging. The development of a clinical prediction model requires a collaborating team of experts in statistics, well qualified in the specific methodology, computer science, researchers and physicians to identify key elements of the model [49]. Predicting outcomes in everyday practice can help guide clinical care, decision making, and resource allocation. In fact, prediction of CKD progression and AKI were rated the highest by participating physicians. Several prediction models have been developed for CKD before, however, are currently infrequently used [50]. Reasons could be that few of them have been appropriately externally validated and out of 23 models only 2 met the proposed criteria [51, 52] for clinical usefulness [50]. The best-known model, the Kidney Failure Risk Equation by Tangri et al. [12], uses 8 routine laboratory parameters. Zacharias et al. [11] established a new risk equation based on data from the GCKD study using six routinely available laboratory parameters and validated the algorithm in 3 independent prospective observational cohort studies. There are tools for useful clinical prediction models to guide development and ensure transparent reporting (TRIPOD [53]: Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis).
Acute kidney injury (AKI) is associated with short- and long-term mortality in hospitalized patients (particularly, if patients are admitted to the intensive care unit [54]) and increased risk of major adverse kidney events including CKD progression [10, 55, 56]. Current international guidelines recommend follow-up of all patients with AKI for 90 days [57]. Recently, a study from Sawhney et al. indicated benefits for re-empting readmissions, death, and subsequent CKD progression for follow-up after AKI [58]. Risk-stratifying CKD patients for AKI has potential to facilitate targeted interventions, thus, physicians were second most interested in an AKI prediction model. Overall, data from large CKD observational studies on the relationship between AKI and CKD are scarce. Identifying patients for increased risk can improve disease management.
The risk of CVD events is higher in patients with CKD than in the general population. However, few models exist to predict the risk of CVD events in patients with CKD to date. Although prediction of CVD in general was not queried in this survey, but stroke and PAOD as secondary complications of CVD were included (M3.3, M3.4). In addition to disease etiology and comorbidities, modifiable lifestyle factors of CKD play an important role in primary prevention [59, 60]. Current CKD management guidelines recommend that patients adhere to a healthy lifestyle including diet with low potassium and low salt intake, physical activity, abstain from tobacco use, and limited alcohol consumption [61, 62]. Adherence to healthy dietary patterns is associated with lower risk for CKD progression and all-cause mortality in CKD [63, 64], however, long-term behaviour change of patients remains a challenge in practice. CKD is a silent disease, i.e., it causes few symptoms, which negatively affects the motivation of patients to change their behaviour. Successfully addressing this problem involves ensuring adequate time spent on conceptualization and planning, as well as visualization of goals with regular adjustment involving the patients.
Ethical aspects with regard to CDSS
CDSS and usage of machine learning techniques will lead to profound changes within healthcare. This also gives rise to many ethical questions, since high-quality patient treatment does not only depend on the correct prediction, prevention or improved decision making by CDSS containing a better selection of individual treatments, but also on experience, empathy, and the concrete judgement of physicians [65]. In this survey, nephrologists were not worried about a reduction in trust from patients when applying and using CDSS. This might be due to the fact, that both patients and physicians share the view that the final act of decision making rests with the human actors. CDSS cannot carry out the full process of clinical care, making CDSS supportive systems to be used by physicians not vice versa also in the future [66]. Moreover, advice by CDSS does not necessarily have to be followed but there is a complex interaction between expert system advice and physicians’ own judgment and discretion which affects physicians’ epistemic authority, particularly in situations of “peer disagreement” [67]. Nonetheless, past research endeavors were able to show that the adoption of CDSS based on clinical practice guidelines (= systematically developed statements to assist practitioners and patient decisions about appropriate health care for specific clinical circumstances) promoted adherence to these guidelines and led to improved patient management [68]. In this survey, most nephrologists did not believe in a reduction in clinical abilities by applying CDSS. Whether CDSS will open up more time for physicians to actually talk to their patients by freeing them from time-consuming decision making seems to be less clear and nephrologists were ambivalent in their answers. Indeed past studies were able to show, that rigidity of CDSS can lead to time-consuming data entry processes that physicians will come to resent leading to rejection of the system [69] and potentially affecting the patient-physician-relationship and the quality of clinical care in a negative way. Moreover, manual data entry can lead to erroneous data entries resulting in inaccurate CDSS responses [70]. An optimal CDSS for nephrologists would therefore entail an internal data entry check, which would help in recognizing false data entries and give feedback to the physician entering the data into the application. Other challenges that need to be faced are, amongst others, poor data sets and poorly optimized algorithms leading to possible discrimination of population fractions or erroneous decision making for underrepresented patient groups. In our survey, only~11% of participants were worried that this might be a problem of CDSS. A study by Mitchell et al., by contrast, could show that using CDSS in high-minority hospitals had higher quality improvements when focusing on CDSS adoption [71]. Nonetheless, underrepresentation of minority groups in clinical trials is a well-known problem [72] and may lead to maladjusted algorithms for these groups concerning CDSS-based prediction of adverse events.
Strengths and Limitations
This study has important strengths and limitations. Since we did not have a tracking system to indicate respondents from non-respondents, we could not compare characteristics of those not included in the study. Having two methods to receive answers from participants, namely telephone interview or do-it-yourself, might also have introduced bias as participants might have been less likely to not give answers for some of the questions. Our study, however, has multiple strengths. This is the first survey of nephrologists exploring their CDSS knowledge and requirements to contemporary needs across Germany. We examined readiness in terms of infrastructure availability and interest to test the instruments for use in their own practice.