Substitution of physicians by nurses is often discussed and widely practiced in many countries, with the aim of satisfying the demands of an aging population and (local) shortages of physicians. Our review showed that the volume of rigorous evaluations is slowly increasing but remains low. In addition, the quality of available research does allow strong recommendations for practice and policy, despite previous proposals [6, 7].
In the appraised literature, the nurses assessed a wide variety of conditions and performed various tasks, with different degrees of clinical autonomy and in different settings. Despite this heterogeneity and the substantial methodological limitations, our review suggested that nurse-led care is associated with higher patient satisfaction, lowered overall mortality and lowered hospital admissions. Effects on other outcomes, such as QoL and costs remained inconclusive.
The effect of nurse-led care on hospital admissions and mortality was particularly present in studies of on-going care and non-urgent visits and when nurse practitioners (both NP and NP with higher degree/courses) provided the care. This suggests that trained nurses can effectively provide healthcare to patients with established diseases. However, the effect disappeared (for hospital admissions) or weakened (for mortality) in studies with better or adequate concealment of allocation and in larger studies. The reasons for this surprising and important finding, especially that nurse-led care could lead to reduced mortality, should be addressed in future studies.
Our overall results also showed a highly significant effect of nurse-led care on patient satisfaction although with severe heterogeneity between trials. This finding is consistent with previous reviews [3, 4]. Nevertheless, this result should be interpreted with caution. Although the average effect is positive, subgroups of patients reported less positive views. Our results suggest this variability may be due to nurses’ roles or study size, which may be associated with other factors (such as degree of clinical autonomy). The effect disappeared when we considered only the trials based on on-going care or non-urgent care, and in trials with longer follow-up episodes (at least 6 months), but these subgroups included two trials only. Surprisingly, patient satisfaction was higher with general nurses (as compared to NPs or NPs with higher degree/extra courses), but the two very small studies showing this effect addressed tasks for very special conditions such as incontinence and family planning. This finding fits in with previous research which showed that patients appreciate nurses’ involvement especially in education and counselling [47, 48].
The results on QoL were difficult to interpret due to heterogeneous reporting of outcomes and the data that were scattered across different scales with outcome measurements at variable follow-up time intervals. Only a few trials used both generic and disease-specific scales with primarily one trial per scale. There was a potential increase in QoL scores with nurse-led care, when health status was evaluated using generic scales, or for specific conditions (e.g. heart failure, Parkinson’s Disease) but the effect was not significant or not sustained at length (at least 12 months) or it was contradicted by data from the same studies . Similarly, there were some effects of lower costs with nurse-led care, but the reported data used different approaches to value the resources and to calculate costs in only a few trials and economic evaluations.
Methodological appraisal of included studies
We identified several significant limitations in the current evidence which should be considered in future research. The trials included were highly heterogeneous in terms of tasks, settings, collection and reporting of outcome measurements. There is a considerable amount of data that are reported in descriptive accounts only, limiting both their pooled validity and the interpretation of their results. Additionally, many studies failed to report some important statistical information (e.g. sample sizes, mean scores, SDs) required to calculate trial estimates and to integrate them in a meta-analysis.
No study fulfilled the set of methodological quality criteria assessed, despite widely available guidelines for RCTs. Trials of lower methodological quality (small study, at least 20% attrition and lack/unclear allocation concealment) tended to inflate the results and only less than 50% of the trials maintained the least target sample required to achieve power, which makes results less trustworthy. The most probable small study bias affecting the effect sizes are the results of small negative studies which are generally less likely to be published than small studies with positive results (i.e. publication bias). Blinding (clinicians, patients and outcome assessors) was reported in only a few trials and we don’t rule out the possibility that patient satisfaction, a subjective outcome, may have been especially positively affected by this. The trials consisted of follow-up episodes of variable length (0.5 to 122.4 months) which may have limited the true effect of care especially in multi-morbid or serious illnesses. Our analyses partly explained the reasons for heterogeneity where this was present but several other variables, which we could not account for, may have also caused this. Patients’ perception and evaluation of satisfaction may be inherently subjective due to socio-demographic differences, experiences from previous care, the physical environment, and patient-care provider interactions. Therefore, measurements of outcome using validated tools are preferred. Of the trials appraised, less than 50% used validated questionnaires for patient satisfaction.
We also identified a lack of trials of cluster randomisation. Although these may be more complex in design, if accounted for all key factors including clustering effect, appropriate sampling and analyses, cluster RCTs could add important value to the current evidence.
Surprisingly, there is a dearth of economic data. The little evidence available on the cost of physician-nurse substitution relies on results which are mainly based on direct costs and use variable approaches. The more recent literature reports more economic data, but it seems difficult to integrate these results especially because cost evaluations differ across countries and thus in cost measurements. We found only two publications [31, 38] providing economic data related to three of the included trials. Despite continued claims of substituting physicians by nurses based on healthcare costs, the evidence can only suggest that substitution is cost neutral. Therefore, as suggested in a recent systematic review of economic evaluations , to meaningfully place the costs and consequences of substitution in the context of healthcare, studies should address all types of costs. Relevant and appropriate data should be generated by means of a systematic collection of economic measures, and specific rules for cost data estimations should also be defined and followed.
More intensive implementation could enhance the outcomes of nursing care, but most studies do not provide the necessary information. In the evaluated studies, the assumption is that nurses possess the competence required for substituting physicians, but the level of substitution does not seem equal among studies. While the level of training may be a critical factor for an effective outcome, the studies report incomplete descriptions of nurses’ roles and competencies. The level of clinical autonomy in nurses does not seem consistent with the level of training and the tasks performed. Also, nurses still require support or communication with the physician for various tasks. It seems then that the level of qualification and training required to carry out substitution requires yet a better definition of practice boundaries including a classification of tasks. Better criteria conceptualised to define nurses’ roles and responsibilities are needed. In addition, the various differences between countries’ definitions and their organisation of nurse care should be taken into account. Lastly, more than half of the evidence reviewed (62.5%) has been conducted in Europe, mainly the UK and the Netherlands.
It is apparent that there is much room for primary studies that include larger numbers of patients, methodologically more rigorous in terms of quality, comprehensive in terms of data and statistical methods and with longer follow-up episodes. Furthermore, in order to gain a better understanding of substitution, future research should map a wider range of nurses, the various levels of training and clinicians’ characteristics, which are provided in many countries. As suggested previously , each method of skill-mix may have its own strengths and weaknesses. The implementation of methodologies aiming at the standardisation of skill-mix studies could support a sound assessment such that health sector reform may also benefit from the publication of evidence.
Strengths and limitations of the review
Our review updates and extends earlier systematic reviews [3, 4] and benefits from a thorough assessment of RCTs, in which the nurse acted as the main figure of care. It also presents (where available) the results by nurses’ roles. Having used the fixed effect model, we can only make inferences about the studies included in the meta-analyses performed here. We only included RCTs because these are at a lower risk of bias and allow for the identification of causal relationships. Although non-randomised trials may overestimate the benefits of nurse-led care it would be recommended to scrutinise the current evidence with such designs. These may not only provide an opportunity for an update but also allow for the collection of data from long term (more than 12 months) follow-up designs which may consist of larger sample sizes. We only included publications in English. We did however screen the reference lists of relevant reviews (some in foreign languages) and searched the reference lists of all included studies. We did not contact authors for further information nor did we search for grey literature. A further limitation is that it was often difficult to understand in detail what role and responsibilities nurses had, when substituting physicians. In many cases, they remain embedded in patient care teams that also involved physicians.