A Review of the Cochrane Review of Nurse Practitioners
as a Possible Replacements for Physicians

Phillip Shaffer, MD

In recent years there has been a legislative push, led by the American Association of Nurse Practitioners (AANP) and supported by many business interests, to allow Nurse Practitioners (NPs) to practice medicine without supervision. To promote this agenda, many claim that “100s of papers” in the medical literature show that NPs are “just as good as physicians”. This broad-brush statement is presented to legislators as a truth, without qualification. Of course, the legislators do not actually read the papers referenced to verify their accuracy; they rely on the presumed honesty of those presenting the information.

The literature review that is most referenced in these discussions was produced by the Cochrane Collaborative and authored by Laurant. They produced two reviews, one in 20051, one in 20182.

It is time to closely review the Cochrane Report and the papers it encompasses to determine where truth lies.

The Cochrane Collaborative report is important to look at closely, not just because it has been used to promote NP independent practice (many other papers have been used for this purpose), but primarily because of the astonishing size of their sample. Despite the sample size, there not a single paper that actually addresses the issue the AANP is using this publication to promote: totally independent practice by nurse practitioners in any specialty they wish.

Laurant screened 8885 articles in the world’s literature (including both the 2005 and 2018 versions of the review.) Only eighteen articles in the world literature were deemed to be of high enough quality to include in the review. The remaining 8867 were not. This, in itself, says much about the general quality of this literature.

However, on close look, even these 18 do not answer questions about nurse independent practice. As we shall see, 16 of the 18 were actually a test of physician-led team care, NOT independent NP care (Table 1).

We can reasonably conclude that there are NO papers in the world’s literature that were done well enough to support the conclusion that NP independent practice is safe and effective.

Only 3 of these 18 papers were even in the United States. These three will be separately discussed. The non-US papers cannot be used for the discussion of NP independence in the United States, primarily because we have no knowledge of the level of education of the nurses in the non-US articles.  However, the 15 non-US papers will be analyzed to give the reader a sense for what sort of articles the nursing lobby is using to promote universal unsupervised practice.

Using Cochrane to conflate primary care with all other specialties

The report is entitled “Nurses as Substitutes for Doctors in Primary Care”. While the article is purely about primary care, it has been used to promote independent practice not just in primary care, but in any specialty practice the NP may wish to work. NPs can, and have, gone far beyond what they were intended to be and have entered independent practices in many specialties such as pediatrics, women’s health care, dermatology, endocrinology, addiction medicine, and many others. To gain the right to do this, they have used this review article as part of the justification for being “just as good as physicians”, even though the publication only discusses primary care.

The author of this review says this:

Owing to the fact that many tasks do not necessarily require the knowledge and skills of a doctor, interest in using nurses to expand the capacity of the primary care workforce is increasing.

Inherent in this statement is the affirmation that physicians are more expert than NPs and that there are many tasks that are not appropriate for the lesser-trained NPs.  There are tasks that only physicians should be performing. Laurant indicates that their review will look at small segments of primary care practice that nurses may be able to capably perform. It does NOT say that nurses could entirely replace physicians. Nevertheless, this review has been used as a tool to promote NP independent practice and obtain the right to do virtually anything in medicine.

What physicians do and how that differs from what Nurse Practitioners are taught

What does it take for someone to practice “as well as a physician”? Let’s review the skills required of a physician when interacting with a patient:

  1. History taking
  2. Physical examination
  3. Developing a provisional differential diagnosis1
  4. Determining a plan of testing to narrow down this differential diagnosis, hopefully to one diagnosis
  5. Determining the best treatment plan for the diagnosis, considering the patient’s other illnesses and medications
  6. Monitoring the results of treatment to determine whether it is effective, and if ineffective, to expand the differential diagnosis to include other, less likely possibilities than originally suspected, and begin testing and later treatment for those.

Every physician must develop a mastery of these six skills, and, most importantly, must prove mastery of the skill by demonstration of mastery to a mentor. Physicians must also pass a standardized test designed specifically to test these skills. This is the National Board of Medical Examiners (NBME) exam, the “Step 3” which is administered every year to physicians who are near the end of their first year of supervised practice, often called the “internship year”. A physician will not be licensed until he or she passes this exam, and 98% do so on the first try.

Nurse Practitioners never have to prove competency in clinical situations by passing such an exam.

Nurse Practitioners receive minimal, if any, training in the first four skills, and only partial training in skills 5 and 6. They do have pharmacology classes, though the coverage is not as in-depth as those in medical school. As for number 6, NPs can monitor treatment, but they are not trained in differential diagnosis, and are not trained to expand that differential diagnosis when it may be necessary. In essence, if the treatment is not working, they have no training in how to address that situation.

Any research that would claim to compare the capabilities of NPs to physicians must address adequacy of the NPs in comparison to physicians at every point in this process. It should also evaluate the accuracy of the diagnoses that NPs declare, and the treatments they prescribe, and compare them to what physicians would diagnose and prescribe.


As you will see, the articles Laurant included in this review do not do this. They do not come close. In nearly every case, the nurse practitioners were assigned patients who already had a diagnosis arrived at by a physician and had already been put on a treatment plan. Thus, the first five tasks of a physician were completed before the NP saw the patient, leaving only monitoring of the patient on a treatment plan.

This is the most common flaw: comparing the NP’s ability to follow a treatment algorithm set up by a physician, and, finding no difference, concluding that NP performance is equivalent.

The ability to accurately follow an algorithm is actually the role NPs were educated to fill, and the role they can capably fill. But these articles are no measure of the capability of the Nurse Practitioner to “substitute for a physician” except in the case of roles that, as Laurant notes “do not necessarily require the knowledge and skills of a doctor”.

An analysis of the Cochrane Review

Common issues with the papers included in the Cochrane Review include the following:

  • trivial and inappropriate endpoints
  • inadequate follow up time to discover issues
  • no primary assessment of undifferentiated patients (patients already had a diagnosis and treatment plan established by a physician)
  • most ill patients were excluded, as NPs were not allowed to see very ill patients.
  • there was direct supervision by physicians in nearly all cases
  • no description of the nurse qualifications/education/training in the studies
  • cross contamination, with patients being also seen by physicians, or changing from one group to another during the study, while still being attributed to the original group (“NP treatment” or “physician treatment”)
  • lack of statistical power required to show non-inferiority of the treatment arm (i.e. nurse led care)–When done, several studies failed to meet the required number of subjects to show non-inferiority. Several other studies had enough patients drop out that no statement could be made. In 5 studies the numbers of NPs studied was so low (1 or 2 nurses) that no general statement about all nurse practice can be made.
  • In only 7 of the papers were nurses allowed to see new patients on their own. In all of these cases, the nurses could refer the patient to the physician. Even when they saw the new patients, none of the papers assessed whether the diagnosis or the treatment plan arrived at by the nurse was correct. The fundamental skill required of a physician was never addressed.

Considering each of these points in detail:

The Endpoints:

“Endpoints” refers to the variables the authors examined to arrive at their conclusions.  Ideally, if one wants to support the hypothesis that nurses can do the same tasks as physicians, there would be evaluation of every one of the 6 skills, such as the evaluation of the history and physical exam of the nurses. Every medical student is evaluated for competency in these skills. In contrast, NONE of these papers evaluated the NPs for these skills, nor did they examine the accuracy of the differential diagnosis, the adequacy of the testing plan, or the adequacy of the treatment plan.

The single variable that was most tested was “patient satisfaction”. This is an entirely inappropriate variable to use to judge the adequacy of medical care. It presumes that the patient knows what good care is. Patients do not generally know this. Gross diagnostic or treatment errors can be made without the patient ever knowing they received inadequate or dangerous care.  Patients are more likely to be “satisfied” with the practitioner who spends more time with them developing a more personal interaction. In some cases, NPs were given twice as much time with patients as physicians.

It has been shown that the most satisfied patients have the poorest outcomes, with a 26% higher mortality rate than those less satisfied.

There were other variables which were often tested such as the “SF-36”. This is a standardized questionnaire about generally how well the patient feels.  This “outcome” is entirely subjective, and there is no indication that a high score on this test means the patient received accurate, effective care.

Some papers did use objective tests: measurements of blood pressure, cholesterol, and similar. These are rather trivial endpoints. Not because they do not matter, but because they are very simplistic measurements of health. Their management only requires following an algorithm. They do not have anything to say about whether the patient was accurately diagnosed, or whether new complaints during the period of the protocol were properly addressed. If you take any group that is already on treatment (as these patients uniformly are), and follow them for 6 months or a year, it is very unlikely there would be significant changes as a group over that time period. These trivial and algorithmic variables may have been chosen because they would be highly unlikely to show change, which may have been the intent of the authors.

The follow up time in the protocol

If one wants to determine if proper medical care is being given, the patient must be followed and observed long enough for any errors to be apparent. For example, one of the papers followed their patients for only 1 week. Another was for only 2 weeks. If there were significant errors, they may well not have been apparent at this point. The longest follow-up period was 1 year (Mundinger), but this was in a relatively young population that was also relatively healthy, and one would not expect many incidents in only 1 year, therefore this, too, was an inadequate test of the NPs capability of handling long term care.

Patient selection

In many of the papers, patients were pre-selected. Those who were significantly ill were excluded from the test group.  These patients went to the physicians exclusively. The nurses saw only the “easiest” patients. This is appropriate, if you are testing the question of whether an NP can substitute for physicians in relatively simple situations. However, this review has been used to promote NPs practicing well beyond this scenario, and that is not appropriate.

The Nurse Qualifications

Several of these papers (Mundinger, Spitzer, Dierick-Van Doyle) had nurses in the NP arms of the study who had clinical education far beyond that of the typical Nurse Practitioner.  In one case, the NPs in the study had 2 years of additional training. Equating the NPs in these three studies to NPs without such training (which is nearly all), is highly misleading. We know that Nurse Practitioner education has drastically changed over the past few decades, for the worse, in most cases, and even leaders within the Nurse Practitioner world are aware and concerned2,3.

In addition, in three papers, only one nurse was in the study, and two papers only included two nurses in the study. These papers are not a test of the general abilities of all current NPs, but rather a specific test of the abilities of these few NPs. Perhaps they can be seen as pilot studies, but certainly no general conclusions about all NPs can be drawn from studies that have 1 or 2 NPs in the trial.

Physician oversight/Contamination of groups.

Physician oversight refers to physicians being available for immediate consultation or being required to sign off and assume legal responsibility for the patient.

Contamination of the groups means that in some or many cases, patients assigned to the NP group were also treated by physicians and vice versa. Unless the investigators prohibited such crossover, the groups would not be a pure measure of the capability of that practitioner; the results become some mix of NP + physician. Most of these papers had this issue.

This demonstrates the practical difficulty of constructing a real test. Suppose you wanted to devise a perfect test of NP vs physician care. You would assign equal numbers of patients to each group but would forbid physicians from interacting with the NP patients. In this hypothetical trial, the NP will at times be unsure of what to do with a patient, but would not be allowed to consult. Obviously, this is against the best interest of the patients, and no Institutional Review Board would permit this.

An additional comment about nurse-physician interaction in these papers:
Most of these studies were actually trials of physician-led teams, just as the role of the nurse was intended and precisely what Physicians for Patient Protection advocates. This is safe and effective care for patients.

These papers stand as an endorsement of the physician-led care team model. They have nothing at all to say about independent nurse practice.

It is striking that AANP, or anyone, uses these to promote legislation for unsupervised care, because they were specifically designed as tests of supervised care. Indeed, one author (Venning) wrote explicitly in their paper “”We have evaluated care given by nurse practitioners working as part of primary care teams alongside general practitioners. Our results do not therefore relate to nurse practitioners who are working independently.” 

The proponents of unsupervised care are using papers that are tests of SUPERVISED care and misleading people by saying these papers support unsupervised care. The opposite is true – they show can fill the role their education was designed for – physician extender – capably.  There is NO test of unsupervised care in any of these papers.

Characterizing these papers as supporting unsupervised care is dishonest.

Statistical Treatment

The goal of most of these papers was to show that NP care is as good as physician care in the chosen clinical situation, be it telephone triage, or follow-up of patients in a clinic. Showing “as good as” is tricky in statistical methods. Some authors publish a “p” value that is greater than 0.05 and take that as evidence of equivalence. This is wrong.

The proper way to do this is to assume (in the parlance of statistics) inferiority, and using statistical methods, disprove inferiority, thus inferring “non-inferiority”5. This is done with a power calculation. The results of this calculation can be stated as “we found a sample size of 1000 would be needed to show a difference of 5% with 80% confidence”. Then, if you have 1000 samples and the results show less than 5% difference, you can say with 80% confidence, that they are the same.

For example, if I want to show that temperatures in April and May are the same, I may sample temperatures on two dates. If we sample 2 days in each month and show that these are not different by some statistical test, we cannot say that the temperatures are no different in April in May. It is intuitively obvious we do not have enough data and acquiring more samples would perhaps show the difference. What is missing is a power calculation that would tell us how many days we would have to sample to show a 4-degree difference with 90% likelihood.

This is but one of the errors found in the papers here. Another is attrition bias. As a guideline, it is typical to require at least 80% follow up of patients in a study to be sure the study is valid. As a thought experiment, it is possible 20% of the patients died, and excluding them would be an obvious error. Some of the papers here (example: Mundinger) did perform a power calculation, did have sufficient sample size, but then had only 79% follow-up at 6 months, and only 55% at 2 years.  No conclusions can be arrived at in this situation. (Worse – at the 2-year point, only 12% were still being treated by the service to which they were assigned at the outset. Anything could be happening to these patients, and ascribing it to the NP or physician assigned originally is an obvious error.)

Another statistical error found in 5 of these papers is rather unusual, as it is almost never seen in literature comparing different treatments. In 3 of these studies, there was only 1 nurse in the “nurse” arm of the study (Chambers, Chan, Lewis). These studies were only a test of those individual nurses and cannot be generalized to all nurses or nurse practitioners. In two other studies, there were only 2 nurses involved. (Spitzer,Houwelling)

Studies specifically about practice in the United States

As noted above, a focus of this discussion needs to be US studies, since Laurant’s review is used to promote legislative agendas in the state legislatures of the US states. There were only 3 of the 9000 papers that addressed the situation in the US. Moreover, these were ALL old studies, the youngest being published in 2000 (the data were gathered in 1995).

Why is this important?

In the US in particular there has been a dramatic change in the education of NPs since that time, with the proliferation of NP schools, many of which are online, and some of which accept 100% of applicants.

This situation is so serious, that even proponents of NP care refer to these as “diploma mills6”. Experienced NPs have been very upset at the degradation of the education of NPs. One survey of over 400 experienced NPs found that 82% felt that new graduates were not prepared to take care of patients7. Studies as old as these do not take this crisis in NP education into account. Laurant’s most recent review was published in 2018, but there has been no quality study done on a US population since 2000. In addition, one of the US studies (Mundinger) was so seriously flawed that it must be disregarded. (See below)


AANP and business interests are pushing for fully independent practice for nurse practitioners, meaning they would be allowed to perform all 6 skills of physicians without supervision. Nurse Practitioner training and education does not encompass all of these skills. In some areas, NPs have no training at all.

AANP cites the Cochrane Collaborative review with this statement: “Overall, health outcomes, and outcomes such as resource utilization and cost were equivalent for nurses and physicians”. They ignore that this conclusion was for narrow and sharply circumscribed duties in primary care. They allow the implication that this is for all specialties and all practices to exist in the readers mind. Then they extend the statement, using it to push an agenda of allowing unsupervised care by NPs. It is important to note that not one of these is about unsupervised care. Sixteen feature the practice model that physicians for Patient Protection endorses – NPs and physicians practicing side by side with ample ability for the NP to consult the physician during the patient visit.  

Review of the individual papers included in Cochrane: United States papers

Hemani 1999, United States

This was primarily a study of resource utilization. It did not evaluate the accuracy of diagnoses or treatment. NPs presented every patient to the attending physician for the first 6 months; thus, this is not a pure measure of NP performance, but of NP performance in a physician-led team.

Nurses ordered 2.46 more CT/MR scans, 2.3 times more Ultrasound exams, and 1.5 times more Ophthalmology consultations than attending physicians. (All statistically significant). The CT/MR and Ultrasound exams are among the most expensive diagnostic tests. The authors speculate the orders were higher because of uncertainty: “Nurse practitioners may have encountered more uncertainty in their practice and ordered more of these tests in search of a diagnosis”.

There was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Lewis 1967

United States

This study is over 50 years old; it does not include graduates any more recently than 50 years. It has been recognized by many that the quality of recent graduates is substantially inferior to that of older Nurse Practitioners.

There was one nurse in the study – this is a test of that single individual’s performance. Patients were selected who were “in a relatively stable phase of the natural history of their disease.”  Stable patients, the easiest patients, without any acute issues or confounding factors, were included in this paper. The nurse’s charts were reviewed on the same day by a physician.

The study does not describe the training level of the nurse. It enrolled only 66 low risk patients. It involved the monitoring of chronic disease.  The nurse did not evaluate new patients. While the nurse’s charts were reviewed on the same day as patients were seen, and there was an opportunity to analyze her diagnostic and treatment capability, there was no evaluation of these.

Mundinger 2000,United States

This paper has been promoted by the primary author as the only study with randomized patient selection. The authors indicate there was no day-to-day supervision of these nurses, but the physicians did have to sign off on the nurses work for legal compliance; ultimately the physicians were responsible. The lead author is one of the most vocal proponents of unsupervised care in the United States and failed to report the Nurse Practitioners in this particular study had extensive additional training well beyond that of typical Nurse Practitioners. There is also no way to measure how much physician involvement each patient had.

The nurses in this study were NOT NPs with a standard NP education. They were experienced (8-20 years) and, moreover, had 9 months of “residency-like” training. They thus had far more clinical training than the standard. In this presentation, Mundinger notes that the investigators knew the NPs were inadequately trained after their degrees. This additional training was meant to bring them up to a higher standard than the usual NP so that they could compete with physicians. This was not revealed in the paper, which is a very serious omission, violating the principle that authors of a scientific paper are required to tell the reader every aspect of the experimental design.  The NPs in this study were in no way similar to the NP with standard education now being graduated by NP schools. This fact was revealed in Mundinger’s subsequent book and in her YouTube presentation about the project.

In addition, this report was from 2000, and the data were acquired in 1995. The type of NP that Mundinger started with, and who received significant additional clinical education beyond typical NP education, is not representative of the type of graduate the schools are now producing, a point that Mundinger herself makes in a subsequent paper.

The patients were enrolled after being seen for an acute episode; therefore, they were presented to the study participants with a diagnosis, and this was only a test of maintenance care. No test of assessment of a new patient, differential diagnosis formation, or treatment planning was within this structure.

They did perform a power calculation, but only for the subjective measures derived from patient questionnaires of their perception of their health. There was no power calculation for the objective measures of blood pressure in hypertensives, glycosolated hemoglobin in diabetics, or respiratory measures in those with asthma, thus, the lack of difference between nurse and physician group is of unknown significance.

The patient groups were generally young, average age of 44, generally healthy, and the follow up in the initial paper was short, only 6 months. One does not expect significant differences to be apparent at this interval.

There was follow up on only 79% of patients at 6 months, and thus attrition bias could be present. At 24 months, only 55% could be followed up, thus there was significant attrition bias. Moreover, of this 55%, only 12% of the initial group was still receiving care from the assigned practice. Were these patients now all receiving care from a physician? That is not noted.

The primary author, Mundinger, had a potentially significant conflict of interest at the time of this study. She was on the Board of Directors of UnitedHealth Group. This company is the parent company of Optum, which is one of the largest employers of Nurse Practitioners in the US. The stock and stock options she received as compensation for her service were reportedly worth $94million in 2013. This is a very serious conflict. It was not reported in the paper. It calls into question the reliability of every other aspect of this paper.

The nurses did not evaluate new patients, there was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Papers from Outside the United States

Campbell, 2014, United Kingdom

This was a test of ability of nurses to follow an algorithm for telephone triage. It does not have any bearing on the ability of nurses or NPs to assess patients, order tests, arrive at a diagnosis and establish effective treatment.

The most ill patients saw physicians as a result of the triage.

Validity of endpoints: Patient satisfaction endpoints such as “experience of care after the request” and “Overall satisfaction” are not appropriate measures of the quality of care. Health status changes after a single telephone interaction is a highly questionable endpoint. Deaths within 7 days were no different between the calls triaged by nurses or physicians.

The nurses did not evaluate new patients, there was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Chambers 1978, Canada

This is a very old study. It is highly questionable if it is still pertinent. It was a comparison of a single NP and a single physician. This nurse attended a 9-month course in clinical decision making, physical examination, and other clinically oriented skills that NPs are not routinely taught. She was far more educated than the current group of diploma-mill graduates. For these reasons, this study cannot be generalized to large numbers of current graduates.

A positive attribute of the study is that it does appear the nurse saw undifferentiated patients, in the same manner as the family physician. However, the nurse could provide treatment or refer to the family doctor. Thus, there was no pure test of her skills. She always had back up.

The endpoints were not objective but were subjective patient reports of how patients felt. The period was only a year. It was not possible to obtain a power calculation, and therefore claims of non-inferiority cannot be addressed. Attempts to conclude this one nurse’s care were non-inferior to the physicians are inappropriate.

There was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Chan 2009, United Kingdom

This was a test of a single nurse who had been trained in a very specific area: follow-up of patients after gastroscopy who had only minor disease (usually reflux/dyspepsia) or were normal. There was only one nurse in this study.  This person functioned adequately and had greater improvement in symptoms than those who saw a GP. However, far more patients kept their appointment with the nurse; most did not keep the follow up appointment with the GP. Therefore, in fact, the authors do not really know how well the GP patients did.

This study is NOT a study of independent practice. It is a study of one nurse, and in a very restricted clinical setting: mild symptoms of a non-serious disease. As such, it cannot be generalized to NPs seeing undifferentiated patients with multiple, and sometimes serious, pathologies

Dierick Van-Daele, 2009, Netherlands

This paper is unusual in that NPs had authority to diagnose. However, these were all “minor” complaints, and the study did not review these diagnoses for accuracy. The nurses had had 12 years average experience and then had 2 years of post-NP clinical education. These are far more educated NPs than the standard.

All prescriptions and referral requests were reviewed by the physicians.

There was no power calculation, so non-inferiority cannot be inferred.

This paper rather convincingly shows capable practice by the nurses in the following situation: 1) Experienced nurses who then obtained an NP degree and then had 2 years of additional clinical training 2) Minor diagnoses 3) worked closely with physicians, who reviewed every prescription and specialty referral.
This work cannot be generalized to NPs who have less experience and clinical training, who are working independently in primary care, seeing non-minor complaints with no supervision.

There was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Houweling 2011, Netherlands

This was not a test of independent practice. Patients were evaluated by physicians, and those thought to be too ill were not seen by nurses. The nurse’s role was noted as “ongoing care for patients with diabetes mellitus type 2”. Physicians instituted the insulin therapy.

The nurses were following an algorithmIf there were any complaints outside of the algorithm, the patient was seen by a physician.

The endpoints are irrelevant as a test of nurse capability because the patients were overseen by physicians, and the nurse’s role was only to monitor the patients per an algorithm. Further, there was inadequate statistical power to evaluate for any differences, therefore, no conclusions really can be drawn from this work.

The nurses did not evaluate new patients, there was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Iglesias 2013,Spain

Endpoints tested were resolution of symptoms, patient satisfaction, patient perception of the quality of information and care received, patient preference. Three of these four were about subjective patient impressions and are not appropriate measures of quality care.

The patients were selected for “low acuity”, and the nurses had care algorithms to follow.

There were patients seen by both nurses and physicians. It is unclear what happened in the analysis with patients seen by both groups, owing to non-resolution by nurses. Thus, when the patient became too difficult for the nurse, the physicians came in.

The nurses did not evaluate new patients, there was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Larsson 2014,Sweden

In this study, nurses substituted for physicians in half of biannual visits to a Rheumatology clinic. There were a very low number of patients, only 107. Patients were monitored by a nurse, and later by a physician. Contamination was built into the study. These were very experienced nurses, yet they did not function independently. All new patients were seen by a physician, and testing/treatment plan developed by physician. Changes in treatment were made only with the physicians’ input and the physicians were solely responsible for changes in medication.

These patients were seen 2 times a year, and this protocol extended for one year, so there was one physician visit and one nurse visit per patient. It appears the entire paper is based on a single visit with a nurse per patient.  There was no evaluation of the ability of the nurse to recognize and respond to side effects of these biologic treatments or recognize any other medical issues with these patients. This is an example of supervised practice, NOT independent care. Physicians made all the substantive decisions.

The nurses did not evaluate new patients, there was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Lattimer 1998,United Kingdom

This paper shows that telephone triage by nurses with decision support software and physician backup results in no more deaths than if physicians handled the calls. It does not speak to the question of whether nurses can independently practice medicine safely. The nurses did not evaluate new patients, there was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Moher 2001,United Kingdom

Practice physicians set up secondary prevention measures for coronary artery disease, and the study was purely a study of whether nurses could follow this algorithmic plan. There was no evaluation of new patients by nurses. There was no effort to allow the nurses to diagnose patients, design testing programs, or design treatment plans.

The nurses did not evaluate new patients, there was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Ndosi 2013,United Kingdom

This was a well-done study for the question being investigated, which was “after diagnosis and treatment planning by a physician, can the nurse institute the plan as well as a physician?” The statistical analysis was thorough, and the answer to this question is yes. Importantly, these were very experienced nurses with a median 10 years’ experience in rheumatology clinic.

This is yet another study of nurse ability to follow algorithms once the diagnosis and treatment plan are developed by a physician. This is an example of a well-done supervised care model, not an independent care model.

Questions they do not answer are how well nurses could evaluate, diagnose, and set up treatment plans for new patients, nor do they address the question of how well the nurses performed relative to the physicians when new issues arise with the patients, such as complications of treatment, or unexplained new symptoms.

They write that Nurses ordered fewer intramuscular and intra-articular injections, and fewer radiographs. They do not evaluate whether these differences were beneficial or harmful to the patients, i.e. – whether these were the “correct” actions. Further, while the nurses did order fewer of these tests, the differences were within 95% confidence intervals, and therefore statistically insignificant. Concluding that there were differences, while not pointing out that those differences statistically insignificant is misleading. A more proper way to state the conclusion of this paper would be, “There were no statistically significant differences in ordering patterns”.

The nurses did not evaluate new patients, there was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Sanne 2010,South Africa

This was a test of the ability of the nurses to follow physician-initiated protocols, after diagnosis by a physician. In the authors’ words, this study “compared nurse vs doctor-monitored HIV care”.  All patients were managed under South African National Guidelines for HIV treatment and were given standard ART regimens.” The experimental nurse monitoring strategy used doctor-initiated ART monitored by primary health-care nurses. Again, this is the physician-led team model.

The statistics were properly done and found non-inferiority of the nurses vs. the physicians in their endpoints. However, this paper examines a very limited delivery of health care: monitoring AIDS therapy.

The nurses did not evaluate new patients, there was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Shum 2000, United Kingdom

This was a test of whether patients would be satisfied with being seen by nurses for minor complaints. All potentially serious health issues were seen outside of the study by physicians. There were no medically significant endpoints: There was no measure of adequacy of diagnosis or treatment.

The authors say: “This study did not examine the content of the consultations in detail. Although we assessed several aspects of clinical outcome, the study did not have enough power to detect differences in rare outcomes. We are thus not able to make any definitive statements about the absolute safety of a service led by nurses in comparison with care offered by general practitioners”.

This paper has nothing to say about the ability of NPs to perform as well as physicians in caring for patients.

The nurses did not evaluate new patients, there was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Spitzer 1973, Canada

This paper is so old that it is not pertinent to the situation in 2021. In addition, its nurse practitioner-subjects were far better trained for clinical medicine than current nurse practitioners. It is, however, historically important, as it was the first to examine the question of performance of NPs in a clinical setting.

This article describes a practice that used NPs as they were designed to be used, in an environment in which they may see some patients in a closely supervised situation. This article has nothing to say about independent practice.

The nurses in this study were very experienced, and there were only two of them. They received special training emphasizing decision-making and clinical judgment, which distinguishes them from the usual NP student. The choices they had for each patient were to give reassurance, specific treatment, or refer to the physician. There was substantial cross over with 45% of the nurse practitioner patients being seen by the physicians also in the first 8 weeks, and 33% seen by physicians after the first 8 weeks.

What is extremely important about this paper is this: It outlines a strategy for safe and effective nurse practitioner utilization in a primary care practice. The authors say:

“The results demonstrate that a nurse practitioner can provide first-contact primary clinical care as safely and effectively, with as much satisfaction to patients, as a family physician. The successful ability of the nurse practitioners to function alone in 67 percent of all patient visits and without demonstrable detriment to the patients has particularly important implications in planning of health- care delivery for regions where family physicians are in short supply. ”

This is precisely the model of care that Physicians for Patient Protection advocates, as distinct from what is being promoted by AANP and other business interests. Some are using this article to claim that NPs can practice safely when fully independent with no input or supervision from physicians. There is no information in this paper (or others) that tests or supports purely independent practice. Using the article in this way is simply dishonest.

There was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Venning 2000, United Kingdom

In this old study, the authors state: “We have evaluated care given by nurse practitioners working as part of primary care teams alongside general practitioners. Our results do not therefore relate to nurse practitioners who are working independently.

Further “There were no differences in health outcome, although the study did not have sufficient power to detect a difference in rare serious events.”

The follow up period was only 2 weeks, which is inadequate to judge if the care given was effective. The endpoints did not address adequacy of evaluation, or adequacy of treatment, they were primarily satisfaction and economic endpoints.

Nurses had access to physician consultation and did use this in 12% of cases, this is significant cross-contamination. The nurses did not evaluate new patients, there was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests.

Voogdt-Pruis 2010, Netherlands

This study was a test of algorithm-following, after diagnosis and planning was already done by physicians. This is acknowledged as an appropriate role for NPs. However, there was no test of diagnostic capabilities, of diagnostic testing, of interpretation of diagnostic tests, or of the ability to recognize other health issues other than the single issue that the protocol targeted: Cardiovascular Risk Factor modification.

As noted earlier, the “satisfaction” endpoint is inappropriate as a measure of quality of care. The outcomes in the four measured endpoints were no different between the physician group and the NP group, except for total cholesterol. In that case, there was a statistically different result in the two groups after one year, favoring the NPs. However, while statistically significant, it was only 4%, and this degree of decrease is unlikely to have any clinically important benefits. This is an example of a statistically significant difference that has no clinical importance.

Again, this was a test of Nurse Practitioners’ ability to follow the algorithm for risk reduction. It has nothing to say about the more general situation of evaluating and treating undiagnosed patients.

The nurses did not evaluate new patients, there was no evaluation of the nurse’s diagnostic capability or of ability to design testing or treatment plans, or to evaluate the results of those tests


This close review of the papers in the Cochrane review shows them to be very limited in their scope. They only review care provided by nurses in extremely limited circumstances, such as telephone triage, and ability to follow a treatment algorithm after diagnosis and treatment plan development by a physician. Some of them do in fact show nurses are able to perform these limited tasks as well as physicians. This is not surprising. This is what their education was designed to do. This is appropriate use of their skills.

None of them address the 6 skills required of a capable physician, most examine only this algorithm-following skill.

A few of these papers do have the nurses seeing new patients, but not a single paper evaluated how well they did with diagnosing the patient and making treatment plans.

Many promoting unsupervised Nurse Practitioners have used the papers included in the Cochran Review to imply to legislators and others that nurses are capable of doing everything a physician can do. This is wrong, and seriously misleading to the point of dishonesty. Hopefully this review will help put an end to this misuse of the data.


[1] “Differential diagnosis” is the concise phrase used by physicians to describe the process of developing a plausible explanation of the patient’s problem after taking a good history and physical examination.  Expert clinicians (which all medical students are expected to be at the conclusion of their training) can rank this list according to the likelihood of the possible diagnosis and use the resulting list to guide testing programs. In my specialty, radiology, we are expected to use the history, physical exam findings and the imaging findings to provide a ranked listing of possible diseases and suggest further testing. If a candidate can only supply one or two possibilities, that candidate will fail. Truly, formation of a differential diagnosis is the heart of medicine.

[2] Penny Kaye Jensen, past president of the American Association of Nurse Practitioners (AANP) took to social media to deliver a message. She wrote that while the AANP was unable to “call out” specific programs due to potential liability, “I personally think those who teach in these programs should be embarrassed and ashamed … some chose to follow the almighty dollar and have no pride.” She further stated that nurse practitioner schools “need to establish rigorous admission criteria into NP [nurse practitioner] programs and not admit ‘every warm body’,” adding, “It’s a major issue that students no longer need at least 5 years’ experience as a RN before applying to a NP program.” Jensen bemoaned the decline in the quality of nurse practitioner programs and wrote “these subpar programs without clinical placements should has [sic] never been accredited, additional clinical hours are needed and nursing needs to establish the BSN as an entry level into practice.” Incredibly, though she knows these students are poorly educated, she is still a staunch supporter of giving these students unsupervised care responsibilities.

Al-Agba, Niran; Bernard,  Rebekah . Patients at Risk: The Rise of the Nurse Practitioner and Physician Assistant in Healthcare . Universal-Publishers. Kindle Edition.

[3] Lee, Jamie  Survey of 400 + Nurse practitioners on the quality of NP schools. In additional files.