Physician anaesthetists versus non‐physician providers of anaesthesia for surgical patients
Abstract
Background
With increasing demand for surgery, pressure on healthcare providers to reduce costs, and a predicted shortfall in the number of medically qualified anaesthetists it is important to consider whether non‐physician anaesthetists (NPAs), who do not have a medical qualification, are able to provide equivalent anaesthetic services to medically qualified anaesthesia providers.
Objectives
To assess the safety and effectiveness of different anaesthetic providers for patients undergoing surgical procedures under general, regional or epidural anaesthesia. We planned to consider results from studies across countries worldwide (including developed and developing countries).
Search methods
We searched the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, EMBASE and CINAHL on 13 February 2014. Our search terms were relevant to the review question and not limited by study design or outcomes. We also carried out searches of clinical trials registers, forward and backward citation tracking and grey literature searching.
Selection criteria
We considered all randomized controlled trials (RCTs), non‐randomized studies (NRS), non‐randomized cluster trials and observational study designs which had a comparison group. We included studies which compared an anaesthetic administered by a NPA working independently with an anaesthetic administered by either a physician anaesthetist working independently or by a NPA working in a team supervised or directed by a physician anaesthetist.
Data collection and analysis
Three review authors independently assessed trial quality and extracted data, contacting study authors for additional information where required. In addition to the standard methodological procedures, we based our risk of bias assessment for NRS on the specific NRS risk of bias tool presented at the UK Cochrane Contributors' Meeting in March 2012. We considered case‐mix and type of surgical procedure, patient co‐morbidity, type of anaesthetic given, and hospital characteristics as possible confounders in the studies, and judged how well the authors had adjusted for these confounders.
Main results
We included six NRS with 1,563,820 participants. Five were large retrospective cohort studies using routinely collected hospital or administrative data from the United States (US). The sixth was a smaller cohort study based on emergency medical care in Haiti. Two were restricted to obstetric patients whilst the others included a range of surgical procedures. It was not possible to combine data as there was a degree of heterogeneity between the included studies.
Two studies failed to find a difference in the risk of death in women undergoing caesarean section when given anaesthesia by NPAs compared with physician anaesthetists, both working independently. One study reported there was no difference in mortality between independently working provider groups. One compared mortality risks between US states that had, or had not, 'opted‐out' of federal insurance requirements for physician anaesthetists to supervise or direct NPAs. This study reported a lower mortality risk for NPAs working independently compared with physician anaesthetists working independently in both 'opt‐out' and 'non‐opt out' states.
One study reported a lower mortality risk for NPAs working independently compared with supervised or directed NPAs. One reported a higher mortality risk for NPAs working independently than in a supervised or directed NPA group but no statistical testing was presented. One reported a lower mortality risk in the NPA group working independently compared with the supervised or directed NPA group in both 'opt‐out' and 'non‐opt out' states before the 'opt‐out' rule was introduced, but a higher mortality risk in 'opt‐out' states after the 'opt‐out' rule was introduced. One reported only one death and was unable to detect a risk in mortality. One reported that the risk of mortality and failure to rescue was higher for NPAs who were categorized as undirected than for directed NPAs.
Three studies reported the risk of anaesthesia‐related complications for NPAs working independently compared to physician anaesthetists working independently. Two failed to find a difference in the risk of complications in women undergoing caesarean section. One failed to find a difference in risk of complications between groups in 'non‐opt out' states. This study reported a lower risk of complications for NPAs working independently than for physician anaesthetists working independently in 'opt‐out' states before the 'opt‐out' rule was introduced, but a higher risk after, although these differences were not tested statistically.
Two studies reported that the risk of complications was generally lower for NPAs working independently than in the NPA supervised or team group but no statistical testing was reported. One reported no evidence of increased risk of postoperative complications in an undirected NPA group versus a directed NPA group.
The risk of bias and assessment of confounders was particularly important for this review. We were concerned about the use of routine data for research and the likely accuracy of such databases to determine the intervention and control groups, thus judging four studies at medium risk of inaccuracy, one at low and one, for which there was insufficient detail, at an unclear risk. Whilst we expected that mortality would have been accurately reported in record systems, we thought reporting may not be as accurate for complications, which relied on the use of codes. Studies were therefore judged as at high risk or an unclear risk of bias for the reporting of complications data. Four of the six studies received funding, which could have influenced the reporting and interpretation of study results. Studies considered confounders of case‐mix, co‐morbidity and hospital characteristics with varying degrees of detail and again we were concerned about the accuracy of the coding of data in records and the variables considered during assessment. Five of the studies used multivariate logistic regression models to account for these confounders. We judged three as being at low risk, one at medium risk and one at high risk of incomplete adjustment in analysis.
Authors' conclusions
No definitive statement can be made about the possible superiority of one type of anaesthesia care over another. The complexity of perioperative care, the low intrinsic rate of complications relating directly to anaesthesia, and the potential confounding effects within the studies reviewed, all of which were non‐randomized, make it impossible to provide a definitive answer to the review question.
PICOs
Plain language summary
Physician anaesthetists versus nurse anaesthetists for surgical patients
Background
There is an increasing demand for surgery, pressure on healthcare providers to reduce costs, and a predicted shortfall in the number of medically qualified anaesthetists. This review aimed to consider whether anaesthesia can be provided equally effectively and safely by nurse anaesthetists (without medical qualifications) as by medically qualified anaesthetists with specialist training.
Study characteristics
The evidence was current up to 13 February 2013. We found six relevant studies, five of which were large observational studies from the US with a comparison group and with study durations from two to 11 years, and one was a much smaller 12 week study from Haiti. There were over 1.5 million participants in the studies. Information for these studies was taken from American insurance databases (Medicare) and from hospital records. The small study was based on emergency medical care after the 2008 hurricanes in Haiti.
Key results
Most studies stated that there was no difference in the number of people who died when given anaesthetic by either a nurse anaesthetist or a medically qualified anaesthetist. One study stated that there was a lower rate of death for nurse anaesthetists compared to medically qualified anaesthetists. One study stated that the risk of death was lower for nurse anaesthetists compared to those being supervised by an anaesthetist or working within an anaesthetic team, whilst another stated the risk of death was higher compared to a supervised or team approach. Other studies gave varied results. Similarly, there were variations between studies for the rates of complications for patients depending on their anaesthetic provider.
Quality of the evidence
Much of the data came from large databases, which may have contained inaccuracies in reporting. There may also be important differences between patients that might account for variation in study results, for example, whether patients who were more ill were treated by a medically qualified anaesthetist, or whether nurse anaesthetists worked in hospitals that had fewer resources. Several of the studies had allowed for these potential differences in their analysis, however it was unclear to us whether this had been done sufficiently well to allow us to be confident about the results. There was also potential confounding from the funding sources for some of these studies.
Conclusion
As none of the data were of sufficiently high quality and the studies presented inconsistent findings, we concluded that it was not possible to say whether there were any differences in care between medically qualified anaesthetists and nurse anaesthetists from the available evidence.
Authors' conclusions
Background
Internationally there are challenges for the provision of anaesthetic services. Current and predicted shortfalls can be explained by an ageing population, increasing demand for surgery, changes to working hours, migration of anaesthetists, pressure on healthcare costs and in some countries a reduction in the number of medical graduates choosing to specialize in anaesthesia (Egger 2006; Egger 2007; Jordan 2011).
Similar pressures are seen in other fields of health care, resulting in a trend towards the use of a nurse‐led rather than a traditional doctor‐led service, such as in primary care and monitoring of long term conditions. However, the development of similar substitutions within the field of anaesthesia has been met with more resistance (Smith 2005).
With regard to cost containment, there is a substantial difference in the salaries of the two personnel within countries (in the United States (US), for example, the salary of an anaesthetist is approximately double that of non‐physician personnel (Kalist 2011)).
Role of non‐physician anaesthetists
For the purpose of this review, and to avoid confusion, the word 'physician anaesthetist' is used for all personnel who are medically qualified, and 'non‐physician anaesthetist' (NPA) for all those who provide anaesthesia without a medical qualification. This includes a change of terms for discussion regarding some countries, for example, in the US they are normally referred to as 'anesthesiologists' and 'certified registered nurse anesthetists' (or CRNAs), respectively.
There are considerable differences in the organization of anaesthetic teams across Europe and internationally (Egger 2007; Meeusen 2010), where anaesthetics may be administered by physician anaesthetists working alone or as part of an anaesthetic team, or by NPAs who in turn may be working alone or as part of an anaesthetic team (Bacon 2002). Between countries there are also significant differences in the length of training of personnel (Egger 2007; Matsusaki 2011; Meeusen 2010).
Non‐physician anaesthetists (NPAs) in developing countries
Low and middle income countries, with large populations living in rural locations, have few physician anaesthetists with ratios of less than one per 100,000 population. For example, Uganda has approximately one physician anaesthetist per two million population (Dubowitz 2010) as opposed to the UK which has 12,000 per 64 million, that is 1:5000 (Walker 2007). These countries have been using non‐physician personnel to deliver many anaesthetic services, for example, Kenya’s nurse anaesthesia training programme (Newton 2010).
Non‐physician anaesthetists (NPAs) in the US
The US has a long history of using nurses to administer anaesthetics. However, as anaesthesia has developed as a physician specialty there is now a majority of medically qualified anaesthetists and considerable debate exists between the two professional groups regarding roles and responsibilities (Bacon 2002; Gardner 2011; Matsusaki 2011). Kalist 2011 says “there is so much overlap between the work they do that it is not clear whether an MDA (physician anaesthetist) actually does anything that a CRNA (NPA) does not do”. In recent years, changes to state law in the US with regard to Medicare and Medicaid reimbursement allow some NPAs to now practice without supervision from a physician anaesthetist. At present there are 17 states who have 'opted out' and NPAs can practice as such (AANA Fact Sheet). Millions of dollars have been spent lobbying for or against this ruling (Bacon 2002).
Non‐physician anaesthetists (NPAs) in the UK and other developed countries
In other developed countries there is variation in the changing roles and responsibilities of anaesthetic providers. A move in some European countries now sees NPAs able to induce general anaesthesia for American Society of Anesthesiologists (ASA) I and II patients under the indirect supervision of a physician anaesthetist (for example, in Denmark, France, Norway and Sweden), whilst in some countries (for example, Netherlands and Norway) nurse anaesthetists with additional training are also able to give sedation under monitored anaesthesia care (MAC), again under indirect supervision (Meeusen 2010). These countries however continue to resist a move towards unsupervised NPAs. In the UK, the introduction of an anaesthesia physician assistant, now called physician assistant (anaesthesia) (PA(A)), pilot training programme from October 2003 attempted to address the predicted shortfall of physician anaesthetists (Wilkinson 2007). However, there are limits to the responsibilities given to a PA(A) and they are provided with supervision from a physician anaesthetist. After the introduction of PA(A)s, the Association of Anaesthetists of Great Britain and Ireland maintains an opinion that “the highest standards of anaesthesia can only be achieved by a physician‐only service” (AAGBI 2010).
Despite differences in opinion regarding the length of training of NPAs in some countries, the potential benefits of independent practice are evident, particularly in rural areas which attract fewer anaesthetists.
Impact of use of non‐physician anaesthetists (NPAs) on patient care
The debate over the use of NPAs has focused on patient safety and the question of whether different providers deliver equivalent quality and safety to patients.
A systematic review has been carried out by Smith et al (Smith 2004). They identified four articles relevant to the review question, none of which were randomized controlled trials (RCTs). The authors were unable to show any significant difference in the safety of using different anaesthetic providers, however they also concluded that, given the methodological flaws in the available studies, this was not evidence of absence of a difference.
Apart from anxieties over patient safety, there are other factors involved in how far the role of an NPA should be developed, such as threats to medically qualified physician anaesthetists’ professional status, access to training and working practices, as well as the wish to avoid the costly and lengthy interprofessional conflict that exists in the US (Kane 2004; Smith 2005).
Why it is important to do this review
Increasing demands on healthcare systems together with a predicted personnel shortfall and the current emphasis on cost containment make this a timely and important review.
This review updates Smith’s existing review (Smith 2004) and aimed to establish what is known about patient safety when anaesthetics are administered by different personnel. We hoped that this may lead to an increase in confidence in the skills of NPAs within the anaesthetic community and may potentially lead to greater flexibility in team roles, both within and between countries, depending on patient need.
Objectives
To assess the safety and effectiveness of different anaesthetic providers for patients undergoing surgical procedures under general, regional or epidural anaesthetic.
A subsidiary question was to determine whether there are types of procedures or patient groups for which a non‐physician anaesthetist is not appropriate. We planned to consider results from studies within different regions (US, UK, other developed countries and developing countries) initially and then assess whether the results were consistent across regions before combining results.
Methods
Criteria for considering studies for this review
Types of studies
We aimed to include RCTs, quasi‐randomized trials in which the allocation to the intervention was decided by non‐random means (such as alternation, digits in date of birth or other identification (ID) number) and cluster randomized trials.
In the absence of RCTs, we included non‐randomized controlled trials (NRCTs) and non‐randomized cluster trials. We considered all designs of observational studies which included a comparison group, including prospective and retrospective cohort study designs, controlled before‐after study designs, prospective and retrospective case‐control study designs and interrupted time‐series. We did not include descriptive studies without a direct comparison group.
If we had identified any RCTs we planned to consider non‐randomized studies (NRS) separately and not include them in a meta‐analysis.
Types of participants
We included studies of patients of all ages undergoing emergency or elective surgery under general or regional anaesthetic in a hospital setting. We also included patients undergoing obstetric surgery.
Types of interventions
We included studies which compared an anaesthetic administered by an NPA working independently with either:
-
an anaesthetic administered by a physician anaesthetist working independently;
-
an anaesthetic administered by a NPA working in a team which was supervised or directed by a physician anaesthetist.
We have taken into consideration the difference in terminology of anaesthetic personnel between countries, which can potentially lead to confusion (Vickers 2002). Throughout we have used the terms 'physician anaesthetist' and 'non‐physician anaesthetist' (NPA), as defined above. Examples of different names for anaesthetic personnel are given in Appendix 1. Where a study author used an unclear term to describe an anaesthetic provider that we were unable to designate to one of the above categories, we aimed to contact the authors to seek clarification. There are also various terms used to describe the role of the main anaesthetic practitioner within a team. Some NPAs may be described as being 'medically directed' (anaesthetic is performed by an NPA whilst the physician anaesthetist oversees no more than four concurrent procedures) or working 'under supervision' (anaesthetic is performed by an NPA who is directed by a physician other than the physician anaesthetist). We ensured that we followed the above definitions as far as possible, aiming to contact authors for clarification if necessary to avoid misclassification of the intervention and comparison in the studies.
Types of outcome measures
Primary outcomes
-
Mortality within 30 days of anaesthetic
-
Failure to rescue between induction and full recovery ("defined as the rate of death after complications" (Silber 2000a))
-
Anaesthesia‐related complications (including cardiac, pulmonary and central nervous system complications due to anaesthesia all within 30 days of anaesthetic)
Secondary outcomes
-
Other minor anaesthetic complications (such as nausea and vomiting, pain, sore throat, dental damage) within 48 hours
-
Length of hospital stay
-
Cost
-
Patient reported satisfaction
Search methods for identification of studies
Electronic searches
We searched for eligible trials in the following databases: the Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library 2014, Issue 2), MEDLINE (via Ovid) (from 1985 to February 2014), EMBASE (via Ovid) (from 1985 to February 2014) and CINAHL (via EBSCO) (from 1985 to February 2014). We also searched trial registers, www.clinicaltrials.gov and the World Health Organization (WHO) International Clinical Trials Registry Platform (http://www.who.int/ictrp/network/en/), for ongoing trials. We also searched Health Management Information Consortium (HMIC) via Ovid, which includes grey literature.
The search strategies for MEDLINE, EMBASE, CINAHL and CENTRAL are presented in Appendix 2. The search strategy did not include any outcomes and was not limited by study design or publication type. No language restrictions were imposed. On retrieval of studies we assessed any free text terms or MeSH terms for NPAs, and if we had not used them we would have included these in a modified search strategy.
Searching other resources
We identified other relevant systematic reviews in the search and undertook forward and backward citation tracking for key articles. We contacted study authors to ask if they knew of other relevant ongoing or unpublished studies. In September 2013 we searched the following clinical trials databases: ClinicalTrials.gov, the metaRegister for Controlled Trials, and the WHO International Clinical Trials Registry Platform.
Data collection and analysis
Selection of studies
Results of the searches were collated and duplicates removed. All titles and abstracts were screened by two authors (Sharon R Lewis (SRL) and Amanda Nicholson (AN)) to remove studies that were very unlikely to be eligible. A pilot of 100 titles was performed before all titles were reviewed in order to clarify criteria for discarding articles at this stage. We planned to identify potentially eligible RCTs and NRS separately. If no abstract was available but the title was possibly relevant, the full text of the article was obtained. We anticipated that we would need to get more full texts for observational studies as abstracts may not contain sufficient detail to allow classification (Section 13.3.1.3) (Higgins 2011).
When all titles and abstracts were screened, the full texts of potentially relevant titles were reviewed by SRL and AN and the data were recorded on the study eligibility section of the data extraction form. We planned to have separate eligibility forms for RCTs and NRS (copies are included in Appendix 3). The NRS eligibility form used study design features rather than study design labels, based on the tools presented at the UK Contributors' Meeting 2012, which were then modified to suit our review needs. These features fall into four groups: Was there a relevant comparison? How were the groups formed? Were the features of the study described below carried out after the study was designed? On what variables was comparability of groups assessed? These were incorporated into the data extraction form (Appendix 3).
A pilot selection of 10 papers were read by SRL and AN and then the investigators met to compare results and modify the forms as required. SRL and AN then read all potentially relevant papers and met to compare results. We referred any differences that could not be resolved by discussion onto Andrew F Smith (AFS).
Data extraction and management
Data were extracted from eligible studies by SRL, AN and Phil Alderson (PA) using a paper‐based data extraction form (see Appendix 3). This form was reviewed after data from the first three papers had been entered, and modified as required. If duplicate publications from the same study were identified, we created a composite dataset from all the eligible publications.
The following items were included in the NRS data extraction form:
-
methods, to include risk of bias assessments (see below);
-
patient group, to include age, sex, relevant sociodemographics, case‐mix;
-
setting, e.g. rural or urban, country;
-
intervention, to include training, experience and the level of supervision, role and responsibilities of NPA;
-
comparison, to include training and experience of anaesthetist;
-
outcome, to include time points i. measured and ii. reported, unit of measurement;
-
analytic methods including Unit of analysis issues;
-
results, to include missing participants, subgroup analyses, both unadjusted and adjusted results.
If relevant information or data were not available in the paper, we attempted to contact the lead author to request the additional details. Disagreements were resolved by discussion and, if necessary, consultation with AFS.
Assessment of risk of bias in included studies
We anticipated that we would encounter a range of NRS designs. We based our risk of bias assessment for NRS on the specific NRS risk of bias tool presented at the UK Contributors' Meeting 2012 (incorporated into the data extraction form for NRS (Appendix 3)).
The direction and impact of bias across different NRS is dependent on individual study features and hence hugely variable and difficult to predict (Deeks 2003). In this review, selection bias and confounding by indication were of particular concern.
Risk of bias for each domain was judged as high, low or unclear, unless specified otherwise.
Allocation (selection bias).
As part of our assessment of study design we recorded the factors which determined participant allocation to the intervention or control group.
Blinding (performance and detection bias)
Accuracy with which the intervention or control group determined
We recorded whether the study personnel or participants were blinded to the allocation of participants and any other measures taken to ensure that the treatment of intervention and comparison groups were equivalent in all aspects other than anaesthetic provider. In addition we recorded the methods and data used to decide which patients belonged to the intervention or comparison group, with an estimate of the risk of inaccuracy scored as high, medium, low or unclear.
Accuracy with which outcomes assessed
For studies using routine hospital data there may be errors or omissions in recording outcomes, depending on coding practice within the hospital. Failure to rescue rates, which rely on the recording of a complication before death, have been shown to be very sensitive to the completeness of coding of these secondary diagnoses (McKee 1999). If the intervention and comparison groups were in different hospitals with different coding practices this may have a considerable influence on results. We assessed whether the outcome data were recorded with knowledge of the anaesthetic provider group (blinding) and also assessed the risk of inaccuracy of outcome data, scored as high, medium, low or unclear.
Incomplete outcome data (attrition bias)
This largely depends on the accuracy and completeness of outcome data. This relies in part on the accuracy with which outcomes are identified within the dataset, as discussed above, but also on the coverage of the outcome dataset. Important questions include whether all participants were correctly identified and linked to the outcome dataset.
Selective reporting
Registration of protocols and analysis plans is not as common for observational studies as for RCTs and there is scope for the study authors to present results only on outcomes found to be significantly associated with the intervention of interest. This is a particular risk when routine data are used which have considerable scope to study a range of different outcomes. We recorded whether the study authors had published analysis plans or protocols.
Other potential sources of bias
We considered the funding sources for each study and any resulting potential conflicts of interest.
Assessment of control for confounding factors
Important confounding factors for this research question are:
-
case‐mix and type of surgical procedure;
-
patient co‐morbidity;
-
type of anaesthetic given;
-
hospital characteristics.
All these variables are plausibly associated both with participant outcome and with the type of anaesthetic provider and so could account for any observed association between anaesthetic provider and participant outcome.
Using the data extraction form (Appendix 3) for each NRS we:
-
identified the relevant confounders described by the researcher;
-
identified the method for identifying relevant confounders as described by the researchers;
-
scored all confounders, including those not specified by the researchers for
-
the risk of imprecision in measurement of confounder, scored as high, medium, low or unclear,
-
the risk of imbalance in confounder between provider groups, scored as high, medium, low or unclear;
-
identified the method used for controlling for confounding at both the design and analysis stage;
-
judged the risk of incomplete adjustment during analysis as high, medium, low or unclear.
Measures of treatment effect
On the data extraction form we recorded all unadjusted and adjusted effect estimates for all eligible outcomes, with details of confounders included for each estimate. Regression coefficients or analysis of covariance would have been recorded for continuous variables.
In an attempt to control for confounding, we used adjusted rather than unadjusted effect estimates from NRS in the analysis and discussion of study findings. If multiple adjusted estimates were given we used the estimate that included the largest number of our pre‐determined key confounders.
Unit of analysis issues
The intervention or comparison group may be decided or assigned at a hospital level and this needed to be accounted for in any analysis, since these hospitals may differ in many respects other than anaesthetic provider. Incorrect analysis would result in residual confounding in the model or inaccurate confidence intervals. The use of multi‐level or hierarchical models or robust standard errors was recorded and if the appropriate analysis was not reported we planned to contact the authors.
Dealing with missing data
We contacted study authors to request missing outcome data or any other methodological details. Missing outcome data were likely to be more of an issue in RCTs or prospective cohort studies, where data are collected specifically for the study. If we had suitable data, we planned to undertake sensitivity analyses to assess the impact of the missing outcomes using, for example, worst case scenario, last observation carried forward and available case analysis.
Assessment of heterogeneity
We expected to find more heterogeneity between NRS than between RCTs, reflecting differences in study design and scope for bias, as well as intrinsic differences in the intervention (Sections 13.6.2.3 and 13.6.2.4) (Higgins 2011). It has been estimated that heterogeneity leads to uncertainty 5 to 10 times that of a 95% confidence interval (Deeks 2003). If we had comparable results for outcomes in different studies we planned to use a forest plot to display the most adjusted estimates from each study and to use Chi2 and I2 statistics to describe heterogeneity. Study characteristics that may be important include:
-
number of confounders included in models;
-
analysis technique used;
-
type of data collection.
Assessment of reporting biases
Reporting bias and missing studies are a more complex issue for NRS than for RCTs. Registration and publication of protocols for observational studies is not as widespread as for RCTs so it is not easy to identify the finite population of studies to be included. For this research question there is in fact an existential question concerning the definition of an eligible study. Since routine hospital databases are used in many studies, it could be argued that the pool of eligible studies would include all hospitals which utilize a range of anaesthetic providers and have electronic longitudinal health databases. It clearly would not be possible to access all these databases. It is not clear whether the size of a study or direction of effect are likely to be associated with likelihood of publication, given that many hospital studies are very large but of uncertain quality. These uncertainties undermine the use of a funnel plot.
We aimed to include a wide range of studies using a wide search and did not exclude any potentially eligible articles without reference to the full text.
Data synthesis
We did not pool estimates of effects from NRS studies(Sections 13.6.2.3 and 13.6.2.4) (Higgins 2011).
If we had comparable effect estimates we would have displayed the most adjusted results from each study in a forest plot but without a pooled estimate. As the results from studies were too disparate to display together in a forest plot we have used narrative synthesis to summarize the direction, size and consistency of effects across studies.
Results
Description of studies
Summary details of each study are in Characteristics of included studies.
Results of the search
There were 11,985 studies identified from electronic searches, 161 studies from forward citation searching and a further 33 from backward citation searching. No studies were identified from clinical trial databases. Having removed duplicates, a total of 8102 unique titles and abstracts were considered and then a further 169 full texts assessed for eligibility. We were unable to obtain full texts for five studies and these are listed in Studies awaiting classification. We performed data extraction and risk of bias assessment on six studies. For the search flow diagram, see Figure 1.
Study flow diagram.
Included studies
Study design
We found no eligible RCTs. All six studies included in the review were non‐randomized. Five studies were retrospective cohort studies using routinely collected hospital or administrative data from participants in the USA (Dulisse 2010; Needleman 2009; Pine 2003; Silber 2000a; Simonson 2007). One of these US studies also included a controlled before and after component presenting results in certain states before and after they opted out from the requirement that NPA be supervised (Dulisse 2010). These studies were large, all with more than 100,000 participants, and the total number of participants across all five studies was over 1.5 million. The sixth study was a smaller cohort study, with 330 participants, based in emergency medical care after the 2008 hurricanes in Haiti (Rosseel 2010).
Study population
Two studies were restricted to obstetric patients (Needleman 2009; Simonson 2007) but the other studies included a range of surgical procedures (Dulisse 2010; Pine 2003; Silber 2000a) and Rosseel 2010 focused on emergency surgery only. Three studies used US Medicare data to determine anaesthetic provider and so the study population was aged over 65 years (Dulisse 2010; Pine 2003; Silber 2000a). Pine 2003 studied elective cases for selected operations (carotid endarterectomy, cholecystectomy, herniorrhaphy, hysterectomy, knee replacement, laminectomy, mastectomy or prostatectomy). These were selected so that the study population would be homogenous. Dulisse 2010 excluded day surgery cases because of uncertainty in measuring mortality or complications in these patients.
Intervention and comparison groups
Five studies reported data for NPAs working independently (Dulisse 2010; Needleman 2009; Pine 2003; Rosseel 2010; Simonson 2007). Dulisse 2010, Needleman 2009, Simonson 2007 and Pine 2003 reported a comparison group of a physician anaesthetist working independently.
The studies varied in the definition of an NPA working under supervision or in a team. Dulisse 2010 had a comparison group of an NPA working as part of an anaesthetic team and Pine 2003 had a comparison group of the anaesthetic being administered by a 'team' which included a physician anaesthetist and NPA but it was not stated who exactly administered the anaesthetic. Needleman 2009 had three comparison groups of NPAs working in a team or being supervised: ANES ‐ CRNA I if a physician anaesthetist was required at all planned caesarean sections; ANES ‐ CRNA II if the physician anaesthetist was not required at all planned caesarean sections; and MIXED in which the team varied depending on location. In Rosseel 2010 the physician anaesthetist supervised the NPA in the control group. We considered all of these comparisons as a single group of NPA working under supervision or in a team. Silber 2000a had intervention and comparison groups of undirected and directed NPA using Medicare definitions (Medicare Policy 2005). The undirected group included cases where anaesthesia was delivered by the NPA alone or supervised rather than directed by a physician anaesthetist or directed by a non‐anaesthetist physician. Unbilled cases were also included in this group. The comparison directed group combined cases in which the physician anaesthetists had personally performed the anaesthetic and cases in which the NPA performed the case under physician anaesthetist direction. We kept this study as a separate comparison group.
Time period of study
The five studies based in the US used data collected prior to 2005 with the earliest study period being 1991 to 1994 (Silber 2000a) and the latest 1995 to 2005 (Dulisse 2010). For Rosseel 2010 the study period was for 12 weeks in the autumn of 2008.
Outcomes reported
All studies reported mortality. Some studies specified inpatient mortality (Dulisse 2010; Pine 2003). In other studies the time period was not specified but as the data were collected from discharge data we assumed it was in‐hospital mortality (Needleman 2009; Simonson 2007). Silber 2000a reported mortality within 30 days of admission.
Failure to rescue (defined as 30 day death rate in those in whom either a complication developed or who died without a complication being recorded) was reported separately only by Silber 2000a but was included in the list of anaesthesia‐related complications reported by Dulisse 2010.
Four studies reported complications (Dulisse 2010; Needleman 2009; Silber 2000a; Simonson 2007) which were often presented in amalgamated groups and it was therefore not possible to extract data on serious airways complications as we had originally planned. We have used the modified outcome of anaesthesia‐related complications. If study authors divided complications into anaesthesia‐related complications (such as International Classification of Diseases (ICD) 9 668.0 codes for complications from labour anaesthesia in Needleman 2009 and Simonson 2007) or more general complications we used the data on anaesthetic‐related complications. The definition of complications used in Simonson 2007 added other ICD codes from the list of patient safety indicators from the Agency of Healthcare Research and Quality (AHRQ). Dulisse 2010 used a list of seven relevant patient safety indicators to define complications, including failure to rescue. Silber 2000a presented data on a single group of postoperative complications which were not all anaesthesia‐related.
No studies reported data on any of our secondary outcomes of length of stay, cost or patient satisfaction.
Excluded studies
There were eight studies that were given particular consideration before exclusion, as listed in Characteristics of excluded studies. Four of these studies did not report data for NPAs working independently (Charuluxananan 2008; Hoffmann 2002; Leonard 2012; Maaløe 2000). In two studies it was unclear whether the NPA was working unsupervised. We successfully contacted the authors of these studies, one of whom was able to confirm that they were supervised (Faponle 2004) and the other was unable to confirm due to the length of time since the report was published (Fleming 1992). One study provided no analysis of data by provider type (Charuluxananan 2005) and Abouleish 2004 did not have any surgical patients.
Other reasons for exclusion included that the reports or abstracts had the wrong study design; had the wrong comparison groups; or were letters, commentaries or editorials with no primary data reported.
Studies awaiting assessment
We considered the full texts of two studies for which we were unable to make a decision regarding eligibility without further information. We attempted to make contact with the author of Carpentier 2000 to establish whether the NPA was working unsupervised as well as the author of Gadir 2007 to establish denominator values for all the data presented. We are still awaiting responses. We were unable to access the full texts of five further studies and have attempted contact with the authors to request copies. See Studies awaiting classification.
Risk of bias in included studies
Risk of bias assessments were completed for all studies. It was intended that a number scoring system (1 to 5) be used for judgements but in practice this was completed using judgements of high, medium or low.
Allocation
All studies were non‐randomized with the allocation to the intervention group based on location differences or healthcare decision makers or participant preference. All studies were therefore at high risk of allocation and selection bias. There was evidence of differences in case‐mix and co‐morbidity between the intervention and comparison groups. See the 'Assessment of control for confounding factors' section.
Blinding
Performance bias
In NRS, blinding of participants and personnel to the allocation of participants is often impossible. None of the included studies were blinded and no specific measures were taken to make sure that the care of the intervention and comparison groups were equivalent in all aspects other than anaesthetic provider. These studies are therefore at high risk of performance bias. Differences in the hospital facilities are an important potential source of bias, which is discussed in the section 'Assessment of control for confounding factors'.
Accuracy with which intervention or control group determined
The use of routine data for research purposes raises an issue about the accuracy of the data used. Three studies based in the US used Medicare part B (billing data) to assign participants to the intervention and comparison groups (Dulisse 2010; Pine 2003; Silber 2000a ). It can be difficult to be confident about whether, for example, a physician anaesthetist was actually administering the anaesthetic and the studies dealt with this uncertainty in different ways. Dulisse 2010 assigned cases as NPA alone or physician anaesthetist alone if there was a claim for only one anaesthetic provider. Participants were assigned to NPA team anaesthesia if there was a modifier on either the physician anaesthetist or NPA claim indicating supervision or direction of the NPA. In addition, cases with no part B form were assigned to NPA team anaesthesia if the procedure took place in a 'pass‐through' hospital. It is unclear how accurate this assumption is or how many cases were involved so we assessed the risk of inaccuracy as medium. Pine 2003 excluded all cases with missing or ambiguous provider codes and we assessed this study as low risk of inaccuracy. Silber 2000a classified the intervention groups as undirected or directed and the undirected group included a large number of unbilled cases but sensitivity analyses with these cases removed gave the same results. Participants with multiple anaesthetics in one single admission were classed as undirected if during any one day of admission there had been no directed anaesthesia procedures. This may have the effect of assigning complex high risk cases as undirected. We assessed this study as at medium risk of inaccuracy.
In the other two US studies, both looking at obstetric patients (Needleman 2009; Simonson 2007), the intervention or comparison group was assigned at hospital level based on surveys about usual anaesthetic personnel at the hospital. This raises unit of analysis issues if the intervention group has been assigned at the level of a hospital but individual patient outcomes are analysed. We judged both these studies as at a medium risk of inaccuracy.
Rosseel 2010 gave no details of how the data on provider were collected and we assessed this study as at unclear risk of inaccuracy.
Accuracy with which outcomes assessed
Detection bias and the accuracy with which outcomes were determined were judged for each outcome measure.
We assumed that recording of mortality would be complete and unaffected by allocation group. Detection bias was therefore judged to be at low risk of bias for this outcome. All‐cause mortality was not reported so it was not possible to identify deaths related to anaesthetic complications.
Complication recording was less clear and usually relied on coding of discharge data. There were several issues that may lead to inaccuracy. Differentiation of complications from existing co‐morbid conditions may be difficult. The accuracy with which outcomes were determined in Silber 2000a was assessed as unclear for both failure to rescue and complications as new complications were differentiated from existing co‐morbidities only on the basis of timing of the code (co‐morbid conditions coded in the three months prior to admission). Dulisse 2010 gave few details of the data source used to assess complications and was assessed as at unclear risk. It was likely that there are differences in coding practices between hospitals, which is a potential source of bias especially in studies in which the anaesthetic provider was assigned at hospital level. We thought that the recording of complications in Needleman 2009 was at high risk of inaccuracy and bias given that anaesthetic provider status was assigned at hospital level and there was the possibility that the coding of discharge data may vary between hospitals.
Incomplete outcome data
The risk of incomplete outcome data in retrospective cohort studies was in part determined by the accuracy of the data used, as discussed above. A further concern was the coverage of the dataset. These studies reported on large numbers of participants and used different sources of data. None of the studies reported details of how records of the same participant from different administrative databases were linked. Few details were given of the number of missing records or failure to link records. Given the high volume of participants, it seemed implausible that a fully coded discharge abstract or data record was identified for each participant. We assessed all the retrospective cohort studies as at unclear risk of attrition bias. With a study population of 330 in one surgical unit, Rosseel 2010 was assessed as at low risk.
Selective reporting
No a priori protocols or analysis plans were available during a search of clinicaltrials.gov for all studies. There was an unclear potential for reporting bias due the number of possible codes which authors could select for outcome data and analysis.
Other potential sources of bias
Both Dulisse 2010 and Needleman 2009 received funding from the American Association of Nurse Anesthetists (AANA) and this was considered a high risk of bias for these studies. Neither Pine 2003 nor Simonson 2007 received funding for their studies and they were therefore considered at low risk of bias. Rosseel 2010 provided an evaluation of a training programme established by Mèdecins Sans Frontières (MSF) and as all the authors worked for MSF it was assumed that it was at high risk of bias. Silber 2000a stated that the study had been largely self‐funded but that background methodology work had been supported by grants from AHRQ and The American Board of Anesthesiology (ABA). It was unclear whether the support from the American Board of Anesthesiology could have biased the study.
Assessment of control for confounding factors
The table of confounders (Appendix 4) summarizes our assessment of the measures taken by study authors to control for potential confounding factors.
Case‐mix
Two of the included studies were restricted to caesarean sections (Needleman 2009; Simonson 2007) and were assessed as at low risk of imprecision and imbalance. Three other studies (Dulisse 2010; Pine 2003; Silber 2000a) included adjustment for differences in case‐mix in the statistical models. These data were based on Medicare databases and were assessed as at low risk of imprecision. There were often large differences in case‐mix between the physician anaesthetist and NPA cases with more complex cases such as cardiovascular surgery more likely to be physician anaesthetist‐only or team anaesthesia than NPA‐only so we assessed the studies as at high risk of imbalance. Rosseel 2010 reported large differences in case‐mix between provider groups but with one death only in the dataset the authors did not adjust the results.
Co‐morbidity
Pre‐existing conditions which affected the risk of a participant suffering a complication after surgery were considered by five studies (Dulisse 2010; Needleman 2009; Pine 2003; Silber 2000a; Simonson 2007). These data were based on coded data in the discharge summary or in the Medicare or hospital database. None of the studies used other data sources such as free text in hospital notes or independent data collection. The major concern was the completeness of these coded data for underlying conditions or risk factors and the differentiation of existing co‐morbidity and new complications. No studies included smoking as a patient characteristic. Dulisse 2010 included only age, sex and race, and no other individual data, and we judged this study to be at high risk of imprecision. Silber 2000a adjusted for 27 coded patient characteristics but only if these codes were used in the three months before admission. We considered that many underlying conditions would not be coded in this way and therefore assessed this study as at high risk of imprecision. Pine 2003 used codes for principal and secondary diagnoses to assess co‐morbidity. These authors referred to a more complete database to assess which conditions were most likely to be pre‐existing and which were new complications and we assessed this study to be at medium risk of imprecision.
Needleman 2009 and Simonson 2007 used a list of obstetric co‐morbidities and both were assessed as at medium risk of imprecision. Needleman 2010 revised their analysis to include more codes for obesity and hypertension following the commentary by Neuman 2010.
Silber 2000a and Simonson 2007 reported important differences in co‐morbidity between the intervention and comparison groups with NPA‐only cases more likely to be emergency admissions. Dulisse 2010 reported only minor differences in race but used a restricted definition of co‐morbidity. Needleman 2009, Pine 2003 and Rosseel 2010 did not report on co‐morbidity in the different groups.
Type of anaesthetic given
No studies considered this as a confounder. This meant that different provider groups may have used different types of anaesthesia, for example, a spinal rather than a general anaesthetic for a certain surgical procedure, but this could not be assessed.
Hospital characteristics
Four studies considered this confounder (Needleman 2009; Pine 2003; Silber 2000a; Simonson 2007). The data were based on American Hospital Annual Surveys (AHA) (Needleman 2009; Pine 2003; Silber 2000a) or the study authors' own hospital survey, or both (Needleman 2009; Simonson 2007). None of these survey data were independently verified but the studies differed in the number of characteristics included. Studies using AHA included a range of items assessing location, staffing, teaching status and technological sophistication and we assessed these as at medium risk of imprecision. Simonson used a more limited, variable set including only size, urban or rural location and teaching status and we assessed this study as at high risk of imprecision. All these studies reported important imbalances, with NPA‐only cases more likely to be based in rural, smaller hospitals with fewer facilities, and were judged as at high risk of imbalance.
Dulisse 2010 analysed data from many different hospitals across the USA but did not adjust for hospital characteristics.
Rosseel 2010 was based in a single surgical centre.
Analysis method
Multivariate logistic regression models were used in five studies (Dulisse 2010; Needleman 2009; Pine 2003; Silber 2000a; Simonson 2007). Rosseel 2010 reported one death only and did not present adjusted results. Dulisse 2010, Pine 2003 and Silber 2000a gave methods for model building and we assessed these studies to be at low risk of errors due to adjustment in analyses. Pine 2003 presented indirectly standardized mortality rates for the different anaesthetic provider groups. Expected mortality rates were calculated using procedure‐specific, stepwise logistic regression models. Needleman 2009 gave no strategy for model building and did not report unadjusted rates or numbers of events or denominators to assess model fit and was assessed as at medium risk of errors due to adjustment in analyses. In Simonson 2007 the rationale for the selection of variables into the final model was not clear. The final model included variables for other labour complications including maternal distress, shock, hypotension and cardiac arrest. We thought that these variables were potentially measures of anaesthetic outcome, or on the causal pathway to anaesthetic complications or mortality, and so they should not have been included in the model. We assessed this study as at high risk of incomplete adjustment in analyses. Simonson 2007 used an appropriate hierarchical model to account for the clustering of intervention data in their analysis and Needleman 2009 adjusted standard errors for clustering within hospitals.
Effects of interventions
Comparison 1: NPA working independently versus physician anaesthetist working independently
Four studies investigated this comparison, two on general surgical patients (Dulisse 2010; Pine 2003) and two on women having caesarean sections (Needleman 2009; Simonson 2007).
Mortality
All four studies reported mortality in the intervention and comparison groups. Needleman 2009 and Simonson 2007 failed to find a difference in the risk of death in women undergoing caesarean section with anaesthetic given to participants by NPAs working independently compared with those given anaesthetic by physician anaesthetist alone. In Pine 2003 there were no significant differences in mortality between the provider groups in either unadjusted or adjusted analyses. Dulisse 2010 reported adjusted results using anaesthesia by a physician anaesthetist working independently in non‐opt out states as the reference group. The risk of mortality was lower in cases given anaesthesia by NPAs working independently in both non‐opt out and opt‐out states. This difference was statistically significant within non‐opt out states but it was not possible to assess the statistical significance between provider groups in opt‐out states. This study did not, however, adjust for hospital characteristics. See Analysis 1.1 (Table 1).
| Study ID | Study population | Unadjusted results | Adjusted results | Confounders included | ||||
|
|
| Effect measure | NPA alone | Physician anaesthetist alone | Effect measure | NPA alone | Physician anaesthetist alone |
|
| Obstetric patients, caesareans | Risk difference (1/10,000) compared to physician anaesthetist alone | ‐1.45 | 0 | Odds ratio (reference = physician anaesthetist alone) | 0.556 | 1 | Co‐morbidity and hospital characteristics | |
|
|
|
|
|
|
|
|
|
|
| Obstetric patients, caesareans | Events/ total (Risk /10,000) | 4/33,236 (1.20) | 13/101,570 (1.28) | / | / | / | Co‐morbidity and hospital characteristics | |
|
|
|
|
|
|
|
|
|
|
| Surgical |
|
|
| Odds ratio (reference = physician anaesthetist alone in non opt out states) |
|
|
| |
| Non opt out |
| / | / | / | 0.899* | 1 | Case‐mix and co‐morbidity | |
| Opt out ‐before |
|
|
|
| 0.651* | 0.797* | ||
| Opt out‐after |
|
|
|
| 0.689* | 0.788* | ||
|
|
|
|
|
|
|
|
|
|
| Surgical | / | / | / | Events/total (Risk/10,000) SMR – standardised to whole study population | 13/101,570 (46) 1.031 | 604/134,335 (45) 1.039 | Case‐mix, co‐morbidity and hospital characteristics | |
* significant difference reported by study authors P = 0.05
Complications
Three studies reported the risk of anaesthesia‐related complications (Dulisse 2010; Needleman 2009; Simonson 2007). Needleman 2009 and Simonson 2007 failed to find a difference in the risk of complications in women undergoing caesarean section with anaesthetic given by NPAs working independently compared with those given anaesthesia by physician anaesthetists alone. Dulisse 2010, using the cases given anaesthesia by a physician anaesthetist working independently in non‐opt out states as the reference group, failed to find a difference in risk of complications between groups in non‐opt out states. In opt‐out states the pattern varied with odds ratios lower for NPA alone than physician anaesthetists alone before opt‐out but higher after opt‐out, but it was not possible to test these differences statistically. See Analysis 1.2 (Table 2).
| Study ID | Study population | Unadjusted results | Adjusted results | Confounders included | ||||
|
|
| Effect measure | NPA alone | Physician anaesthetist alone | Effect measure | NPA alone | Physician anaesthetist alone |
|
| Obstetric patients, caesareans | Rate difference (1/10,000) compared to physician anaesthetist alone | ‐6.0 | 0 | Odds ratio (reference = physician anaesthetist alone) | 0.732 | 1 | Co‐morbidity and hospital characteristics | |
| Obstetric patients, caesareans | Events/total (Risk /10,000) | 192/33,236 (57.8) | 773/101,570 (76.1) | Odds ratio (reference = physician anaesthetist alone) | 1.046 (95% CI 0.649‐1.685) | 1 | Co‐morbidity and hospital characteristics | |
| Surgical |
|
|
|
|
|
|
| |
| Non‐opt out |
| / | / | / | Odds ratio (reference = physician anaesthetist alone in non‐opt out states) | 0.992 | 1 | Case‐mix and co‐morbidity |
| Opt out ‐ before |
|
|
|
| 0.798* | 0.824* | ||
| Opt out ‐ after |
|
|
|
| 0.927 | 0.818* | ||
* significant difference reported by study authors P = 0.05
CI (confidence interval)
Comparison 2: NPA working independently versus NPA working in a team which is supervised or directed by a physician anaesthetist
Four studies investigated this comparison, three in general surgical patients (Dulisse 2010; Pine 2003; Rosseel 2010) and one in women having caesarean sections (Needleman 2009). Two studies (Dulisse 2010; Needleman 2009) had several comparison groups and results were presented using the physician anaesthetist working independently as the reference group. This meant that it was not possible to assess the statistical significance of the differences between NPA working independently and NPA working under supervision or in a team, but the relative size of the odds ratios gave an indication of whether mortality or complication risk was higher or lower.
Mortality
In Needleman 2009 the risk of mortality was lower in the NPA‐only group than in the NPA supervised or team group. In Dulisse 2010 the pattern varied with the mortality risk lower in the NPA‐only group in non‐opt out states and opt‐out states before opt‐out but higher in opt‐out states after opt‐out. In Pine 2003 the mortality risk was higher in the NPA‐only group than in the NPA supervised or team group but no statistical testing of any of these differences was presented. Rosseel 2010 reported one death only in a study of 330 participants and so no difference in mortality risk was detected. See Analysis 2.1 (Table 3).
| Study ID | Study population | Comparison group | Unadjusted results | Adjusted results | Confounders included | ||||
|
|
|
| Effect measure | NPA alone | NPA supervised | Effect measure | NPA alone | NPA supervised |
|
| Obstetric patients, caesareans | Physician anaesthetist present at all c‐sections | Rate difference (1/10,000) compared to Physician anaesthetist only | ‐1.45 | ‐0.54 | Odds ratio (reference = Physician anaesthetist alone) | 0.556 | 0.708 | Co‐morbidity and hospital characteristics | |
|
| Physician anaesthetist not present at all c‐sections | ‐1.45 | ‐0.61 | 0.556 | 0.716 |
| |||
| Surgical |
|
|
|
|
|
|
|
| |
| Non opt out |
| Team | / | / | / | Odds ratio (reference = Physician anaesthetist alone in non opt out states) | 0.899* | 0.959* | Case‐mix and co‐morbidity |
| Opt out ‐ before |
| Team |
|
|
| 0.651* | 0.708* | ||
| Opt out ‐ after |
| Team |
|
|
| 0.689* | 0.565* | ||
| Surgical | Team | Events/total (Risk /10,000) SMR – standardised to whole study population | 151/33,151 (46) 1.031 | 796/236,708 (34) 0.967 | Case‐mix, co‐morbidity and hospital characteristics | ||||
| Surgical | Physician anaesthetist supervision | Events/ total
| 0/168 | 1/162 | / | / | / | / | |
* significant difference reported by study authors P = 0.05
Complications
Results presented in Dulisse 2010 and Needleman 2009 were similar to those for mortality, with the risk complications generally lower in the NPA‐only group than in the NPA supervised or team group, but no statistical testing was reported. See Analysis 2.2 (Table 4).
| Study ID | Study population | Comparison group | Unadjusted results | Adjusted results | Confounders included | ||||
|
|
|
| Effect measure | NPA alone | NPA supervised | Effect measure | NPA alone | NPA supervised |
|
| Obstetric patients, caesareans | Physician anaesthetist present at all caesareans | Rate difference (1/10,000) compared to Physician anaesthetist only | ‐6.0 | ‐11.0 | Odds ratio (reference = Physician anaesthetist alone) | 0.732 | 0.832 | Co‐morbidity and hospital characteristics | |
|
| Physician anaesthetist not present at all caesareans | ‐6.0 | ‐2.0 | 0.732 | 0.922 |
| |||
| Surgical |
|
|
|
|
|
|
|
| |
| Non opt out |
| Team | / | / | / | Odds ratio (reference = Physician anaesthetist alone in non opt out states) | 0.992 | 1.67* | Case‐mix, co‐morbidity and hospital characteristics |
| Opt out ‐before |
| Team |
|
|
| 0.798* | 0.927 | ||
| Opt out‐after |
| Team |
|
|
| 0.927 | 0.903 | ||
* significant difference reported by study authors P = 0.05
Comparison 3: undirected NPA versus directed NPA
Silber 2000a presented data for this comparison, in which the intervention undirected group included cases where anaesthesia was delivered by NPA alone, or a NPA was supervised rather than directed by a physician anaesthetist, or a NPA was directed by a non‐anaesthetist physician. The comparison‐directed group combined cases in which the physician anaesthetists had personally performed the anaesthetic and cases in which the NPA performed the case under physician anaesthetist direction.
There was some evidence that the risk of mortality and failure to rescue was higher in the undirected NPA group, with adjusted odds ratios (OR) of 1.08 (95% confidence interval (CI) 1.00 to 1.15) and 1.10 (95% CI 1.01 to 1.18), respectively. In adjusted analyses there was no evidence of an increased risk of postoperative complications in the undirected group. However, the unadjusted ORs were higher for mortality (OR 1.35, 95% CI 1.26 to 1.44), failure to rescue (OR 1.15, 95% CI 1.08 to 1.24) and complications (OR 1.31 95% CI 1.28 to 1.45). Adjustment for differences in case‐mix, co‐morbidity and hospital characteristics accounted for much of the observed increased risk in outcomes. We assessed that co‐morbidity had a high risk of imprecision and the remaining increased effect seen may have been due to residual confounding. See Analyses 3.1, 3.2 and 3.3 (Table 5).
| Study ID | Study population | Unadjusted results | Adjusted results | Confounders included | ||||
|
|
| Effect measure | Undirected NPA | Directed NPA | Effect measure | Undirected NPA | Directed NPA |
|
| Surgical | Odds ratio (reference – directed NPA) | 1.35 (95% CI 1.26 to 1.44) | 1 | Odds ratio (reference – directed NPA) | 1.08 (95% CI 1.00‐1.15) | 1 | Case‐mix, co‐morbidity and hospital characteristics | |
CI (Confidence Interval)
Discussion
Summary of main results
This review included six non‐randomized studies (NRS) evaluating clinical outcomes when physician anaesthetists are compared with non‐physicians, either working alone or in teams of various combinations. Overall, while some studies have shown small and inconsistent differences in some outcomes, the quality and nature of the evidence are insufficient to draw firm conclusions about relative benefits and risks of the different models of anaesthetic provision. Perioperative risk is composed of three elements, the patient’s pre‐existing condition (for instance, the risk of pulmonary aspiration of gastric contents (Smith 1997)), the operation performed, and the perioperative care received, of which anaesthetic care is only one part. The included studies have not been able to successfully separate anaesthetic care from other risk factors.
Overall completeness and applicability of evidence
The included studies were mainly from the United States (US) and used routinely‐collected administrative data. Only one study was carried out in the developing world. We found none from countries with advanced healthcare systems outside the US. Within the US, the data presented may not be representative as they may be skewed to the more deprived. Further, only billed cases were included, which raises the possibility of a systematic bias in the coverage of the data.
It is important to be aware of potential biases in the studies themselves. In the US, there are tensions between the official positions of the two professional organisations of the two main groups of anaesthesia providers, physician anaesthesiologists and registered nurse anaesthetists (Kane 2004). Some of the studies included in this review were funded, at least in part, by those professional organisations and were published in their own journals. Whilst this does not invalidate the results, it is unlikely that one group would publish work which weakened its own political position. The nature and small number of the studies included made it impossible to apply the usual methods used to detect publication bias (for instance, funnel plots) and this has to remain a possible source of bias.
Quality of the evidence
All studies were non‐randomized and so are at considerable risk of bias due to the effects of confounding and selection bias. Further, as different studies took different approaches to definitions and adjustment, it was not possible to compare them directly. It was also problematic trying to fully control for differences in hospital characteristics, and especially for patient co‐morbidity, using these routinely‐collected data. It is open to speculation in which direction such confounding factors might be operating; in general, common sense would dictate that the skill levels of anaesthetic providers would be matched, where possible, to the complexity and riskiness of the patient’s condition and the surgical procedure. As Needleman notes (p465), "The model of anaesthetic provision may be a proxy for other clinical resource variables usually left unmeasured in typically used databases" (Needleman 2009). Finally, no study assessed cost, length of hospital stay or the patient’s perspective as an outcome.
Potential biases in the review process
We conducted a comprehensive search for material, including what is sometimes termed ‘grey’ literature (non‐peer reviewed reports etc.). However, we did not access the grey literature database National Technical Information Service (NTIS), which may have potentially included more American literature.
We considered each eligible study with three independent reviewers and took time to understand the complexities of the American healthcare insurance system on which most of the studies were based. We sought advice from American peers, where necessary, and requested additional information from study authors who mostly responded promptly to our requests. Despite this, it is possible that our lack of intimate understanding of the American healthcare insurance system may have biased our interpretation of the included studies.
Agreements and disagreements with other studies or reviews
The only other review in this area which we are aware of is that of Smith (one of the authors on the present review) and colleagues from 2004 (Smith 2004). The present review has identified four studies published since 2004, and also excluded two studies which the 2004 review included. In addition, we used more recently developed, more sophisticated techniques for assessing risk of bias in non‐randomized studies. There is little difference in the conclusions between the two reviews.
| Study ID | Study population | Unadjusted results | Adjusted results | Confounders included | ||||
|
|
| Effect measure | NPA alone | Physician anaesthetist alone | Effect measure | NPA alone | Physician anaesthetist alone |
|
| Obstetric patients, caesareans | Risk difference (1/10,000) compared to physician anaesthetist alone | ‐1.45 | 0 | Odds ratio (reference = physician anaesthetist alone) | 0.556 | 1 | Co‐morbidity and hospital characteristics | |
|
|
|
|
|
|
|
|
|
|
| Obstetric patients, caesareans | Events/ total (Risk /10,000) | 4/33,236 (1.20) | 13/101,570 (1.28) | / | / | / | Co‐morbidity and hospital characteristics | |
|
|
|
|
|
|
|
|
|
|
| Surgical |
|
|
| Odds ratio (reference = physician anaesthetist alone in non opt out states) |
|
|
| |
| Non opt out |
| / | / | / | 0.899* | 1 | Case‐mix and co‐morbidity | |
| Opt out ‐before |
|
|
|
| 0.651* | 0.797* | ||
| Opt out‐after |
|
|
|
| 0.689* | 0.788* | ||
|
|
|
|
|
|
|
|
|
|
| Surgical | / | / | / | Events/total (Risk/10,000) SMR – standardised to whole study population | 13/101,570 (46) 1.031 | 604/134,335 (45) 1.039 | Case‐mix, co‐morbidity and hospital characteristics | |
| * significant difference reported by study authors P = 0.05 | ||||||||
| Study ID | Study population | Unadjusted results | Adjusted results | Confounders included | ||||
|
|
| Effect measure | NPA alone | Physician anaesthetist alone | Effect measure | NPA alone | Physician anaesthetist alone |
|
| Obstetric patients, caesareans | Rate difference (1/10,000) compared to physician anaesthetist alone | ‐6.0 | 0 | Odds ratio (reference = physician anaesthetist alone) | 0.732 | 1 | Co‐morbidity and hospital characteristics | |
| Obstetric patients, caesareans | Events/total (Risk /10,000) | 192/33,236 (57.8) | 773/101,570 (76.1) | Odds ratio (reference = physician anaesthetist alone) | 1.046 (95% CI 0.649‐1.685) | 1 | Co‐morbidity and hospital characteristics | |
| Surgical |
|
|
|
|
|
|
| |
| Non‐opt out |
| / | / | / | Odds ratio (reference = physician anaesthetist alone in non‐opt out states) | 0.992 | 1 | Case‐mix and co‐morbidity |
| Opt out ‐ before |
|
|
|
| 0.798* | 0.824* | ||
| Opt out ‐ after |
|
|
|
| 0.927 | 0.818* | ||
| * significant difference reported by study authors P = 0.05 CI (confidence interval) | ||||||||
| Study ID | Study population | Comparison group | Unadjusted results | Adjusted results | Confounders included | ||||
|
|
|
| Effect measure | NPA alone | NPA supervised | Effect measure | NPA alone | NPA supervised |
|
| Obstetric patients, caesareans | Physician anaesthetist present at all c‐sections | Rate difference (1/10,000) compared to Physician anaesthetist only | ‐1.45 | ‐0.54 | Odds ratio (reference = Physician anaesthetist alone) | 0.556 | 0.708 | Co‐morbidity and hospital characteristics | |
|
| Physician anaesthetist not present at all c‐sections | ‐1.45 | ‐0.61 | 0.556 | 0.716 |
| |||
| Surgical |
|
|
|
|
|
|
|
| |
| Non opt out |
| Team | / | / | / | Odds ratio (reference = Physician anaesthetist alone in non opt out states) | 0.899* | 0.959* | Case‐mix and co‐morbidity |
| Opt out ‐ before |
| Team |
|
|
| 0.651* | 0.708* | ||
| Opt out ‐ after |
| Team |
|
|
| 0.689* | 0.565* | ||
| Surgical | Team | Events/total (Risk /10,000) SMR – standardised to whole study population | 151/33,151 (46) 1.031 | 796/236,708 (34) 0.967 | Case‐mix, co‐morbidity and hospital characteristics | ||||
| Surgical | Physician anaesthetist supervision | Events/ total
| 0/168 | 1/162 | / | / | / | / | |
| * significant difference reported by study authors P = 0.05 | |||||||||
| Study ID | Study population | Comparison group | Unadjusted results | Adjusted results | Confounders included | ||||
|
|
|
| Effect measure | NPA alone | NPA supervised | Effect measure | NPA alone | NPA supervised |
|
| Obstetric patients, caesareans | Physician anaesthetist present at all caesareans | Rate difference (1/10,000) compared to Physician anaesthetist only | ‐6.0 | ‐11.0 | Odds ratio (reference = Physician anaesthetist alone) | 0.732 | 0.832 | Co‐morbidity and hospital characteristics | |
|
| Physician anaesthetist not present at all caesareans | ‐6.0 | ‐2.0 | 0.732 | 0.922 |
| |||
| Surgical |
|
|
|
|
|
|
|
| |
| Non opt out |
| Team | / | / | / | Odds ratio (reference = Physician anaesthetist alone in non opt out states) | 0.992 | 1.67* | Case‐mix, co‐morbidity and hospital characteristics |
| Opt out ‐before |
| Team |
|
|
| 0.798* | 0.927 | ||
| Opt out‐after |
| Team |
|
|
| 0.927 | 0.903 | ||
| * significant difference reported by study authors P = 0.05 | |||||||||
| Study ID | Study population | Unadjusted results | Adjusted results | Confounders included | ||||
|
|
| Effect measure | Undirected NPA | Directed NPA | Effect measure | Undirected NPA | Directed NPA |
|
| Surgical | Odds ratio (reference – directed NPA) | 1.35 (95% CI 1.26 to 1.44) | 1 | Odds ratio (reference – directed NPA) | 1.08 (95% CI 1.00‐1.15) | 1 | Case‐mix, co‐morbidity and hospital characteristics | |
| CI (Confidence Interval) | ||||||||
