This is EXTREMELY long but worthwhile to read if the topic interests you.
Linda Aiken et. al., in 2003 released through the JAMA a landmark Study, "Educational Levels of Hospital Nurses and Surgical Patient Mortality"
The study is oft cited and contends that evidence suggests that higher rates of BSN education at the bedside directly translates to improved pt outcomes.
This is my critique:
Why the study, "Educational Levels of Hospital Nurses and Surgical Patient Mortality" is flawed.
1. Academic Laziness
The original data pool was used for an earlier study about staffing levels and mortality GENERALLY. That data was just copied onto this template for this study. But it wasn't just copied; it was copied with the full assurance of the authors that the results of the first study that used this data could be 'factored out' of this, subsequent study.
2. Discrimination Bias (Hospital Selection)
Before analyzing the data, the authors first decided that it would be necessary to 'exclude' hospitals that didn't fit their data set. Some were excluded for valid reasons (they didn't report to the data set. Although, however valid, the exclusion ITSELF taints the data. THIS IS EXPECIALLY TRUE SINCE THIS EXCLUSION INCLUDES ALL VA HOSPITALS - a known source for high BSN recruitment. The very hospitals that might yield some useful data on the subject were ELIMINATED from the study!) - but some were excluded because the data they generated didn't meet the authors' needs. In other words, INCLUSION of said data would disturb the conclusions of the study.
So the authors warrant that exclusion of some data is relevant. Ok, I can concede that point as I understand that large standard deviation multiples (outlying data) can skew the majority of data. But, excluding large amounts of data that are quite possibly within a single standard deviation of what is being studied on the basis that such data wasn't available serves the purpose of undermining the whole study. It is a frank admission that the data itself is incomplete, and so, suspect.
This is the compounded error of the academic laziness mentioned above. The data set was copied from another study with the full understanding that it didn't meet the needs of this study, AND COULD NOT MEET THE NEEDS OF THIS STUDY because of its lack of inclusion of hospitals MOST LIKELY to represent a significant sample of this study. Rather than develop data that was 'pertinent' to THIS study, that academic laziness now calls for this lacking, and possibly highly relevant, data to merely be excluded from consideration.
3. Degree Bias
The Authors state in the study, "Conventional wisdom" is that nurses' experience is more important than their educational levels." It is this ''conventional wisdom' that the study aims to examine. But how does it do so? By buying into the exact same conventional wisdom!: "Because there is no evidence that the relative proportions of nurses holding diplomas and associate degrees affect the patient outcomes studied, these two categories of nurses were collapsed into a single category"
HOLD ON. In a study about how degrees affect patient outcomes, an essential tenet of the study is to disregard degrees held??? After such manipulation, how can you say with a straight face that a study that disregards the relationship between degrees can make a conclusion REGARDING the relationships between degrees?
4. Lack of Substantiating Data
"It was later verified that this decision did not bias the result."
This statement, or others like it, appear throughout this 'study' without mention of the methods used to 'verify'.
"Previous empirical work demonstrated. . ." - um, exactly WHAT empirical work was that?
In fact, the study makes lots of claims and manipulates the data in lots of ways that it nevertheless insists that you have to trust its 'independent verification' that such didn't bias the results. Of course, that is without being provided access to said independent verification.
You have to love the 'self-affirming' validity of it all.
5. Data Manipulation
A. The data was 'manipulated' to grant varying degrees of credibility depending upon whether it was received by a 'teaching' hospital vs. a 'non'-teaching hospital.
B. The data was 'manipulated' to grant varying degrees of credibility to hospitals that are more 'technological' (e.g. have transplant services) as opposed to less.
C. "An important potential confounding variable to both clinical judgment and education was the mean number of years of experience working as an RN": telling comment, but never fear, the data was 'manipulated' to take this into account.
D. Nursing workloads might affect patient outcomes. (Indeed, THIS was the previous study that this study's data set was copied from.) But, in this case, the data was 'manipulated' to take those workloads into account.
E. "Estimated and controlled for the risk of having a board certified Surgeon instead of a non-board certified Surgeon." The use of 2 'dummy variables' comparing MD licenses to general vs specialty board certification was "a reasonable way for controlling surgeon qualifications in our models."
In fact the authors admit to manipulating the data 133 ways! But all of these 'manipulations' were later 'verified' to have produced no bias.
6. Key Criteria Conjecture
The study's two key criteria: deaths within 30 days of hospital admissions and deaths within 30 days of complications due to 'failure to rescue'. But how were these criteria established?
In the first case, they were established by comparing the data set to vital statistic records (death records). I doubt they accurately compared 235,000 individual patients (data points) against another data set (death records) that was probably multiple times in size, but OK - I'll buy this for the moment.
In the second case, however, 'failure to rescue' was defined - NOT BY EXAMINING ACTUAL CASES OF FAILURE TO RESCUE - but by establishing different ICD-9 secondary codes from admit to discharge. An assumption is made that a different code meant that a 'failure to rescue' had occurred. What?!
RE-READ THAT LAST! By making dubious assumptions on data sets (hospital reporting statistics) - this study conjectures how this translates to 'failure to rescue' and then makes conclusions based on what this 'failure to rescue' might mean! ALL BY ITSELF, THIS NEGATES THE ENTIRE STUDY.
But, it was 'verified' to not bias the study results. How was this part 'verified'? Well, you're gonna love this: "expert consensus as well as empirical evidence to distinguish complications from pre-existing co-morbidities."
In other words, the experts (the study authors) know which data is valid for purposes of inclusion into the study - and which data isn't. The 'experts' consensus is the key element that ensures non-bias.
There are no 'double blind' studies. No sample populations of RNs. The criteria for inclusion of 'data' is solely based on the 'consensus' of the 'experts' creating the study. And these 'experts': backed by AACN (Amer Assoc of Colleges of Nursing) - An organization committed to BSN-entry and an organization which maintains, on its website, a valiant defense of this study:
http://www.aacn.nche.edu/Media/TalkingPoints2.htm
No, no possibility of bias here.
Let me ask you this: if you know of a study conducted by Republican Pollsters - where they alone determined whose answers were valid - would you trust a result that brags that 'Most Americans Love President Bush!' But here's the question I really want to ask: WHY wouldn't you trust such a result?
7. Risk Adjustment.
Still trust this study? Try this one: "Patient outcomes were risk-adjusted by including 133 variables in our models, including age, sex, whether an admission was a transfer from another hospital, whether it was an emergency admission, a series of 48 variables including surgery type, dummy variables including the presence of 28 chronic, pre-existing conditions as classified by ICD-9 codes, and interaction terms chosen on the basis of their ability to predict mortality and failure to rescue in the current data set."
So the data was manipulated 133 ways, excluding some data. But, and this is key: there are SO VERY MANY variables that could effect patient outcomes that you have to adjust for EVERYTHING except for what you're looking to find. Right? This is not only what the authors contend, but they contend that they SUCCESSFULLY adjusted the data, 133 different ways, for just this purpose, and completely without bias. Amazing.
8. Logistics Regression Models
So, after the study took in all this manipulated 'data', it compared hospitals with higher BSN-RNs to those with less, and reached a conclusion. Right? Wrong.
It took the data and ran a 'logistics regression model' as to what might happen in a given hospital "if there were a 10% increase in BSN RNs."
This study doesn't even compare the relative levels of RN education. Let me repeat that: THIS STUDY DOESN'T EVEN MAKE THE COMPARISONS IT PURPORTS TO HAVE STUDIED. This model and, as a result, this study doesn't compare existing situations. Instead, it makes assumptions regarding potential situations compared to current situations.
Do you get this: the study wasn't designed to test real conditions. The study was designed to create hypothetical situations and comment on the validity of said models based on highly modified and incomplete data.
THIS STUDY SPECIFICALLY COMMENTS ONLY ON HYPOTHETICAL SITUATIONS. Study Disclaimer: ANY RELATIONSHIP TO REAL CONDITIONS IS ONLY IMPLIED BY THE AUTHORS.
Now, see if this isn't a key statement: "The associations of educational compositions, staffing, experience of nurses, and surgeon board certifications with patient outcomes were computed before and after controlling for patient characteristics and hospital characteristics." Indeed.
9. Direct Standardization Models.
Apparently, even after all the above manipulation, there were still 'clusters of data' that had to be 'standardized' using 'robust estimations'. The study does at least have the guts to admit that such 'standardizations' turns the final conclusion into an 'estimation'. Too bad it only makes that admission in the body of the study, and not in the 'abstracts'.
10. Alternative Correlations
The study admits that fewer than 11% of hospitals in Penn in 1999 (the area/year of the study) had 50% or greater BSNs (Excluding the VA hospital system, which were completely ignored by the study.) And then the study cites co-factors that could unduly influence the study under these situations: "Hospitals with higher percentages of BSN or masters prepared nurses tended to be larger and have post-graduate medical training programs, as well as high-tech facilities. These hospitals also had slightly less experienced nurses on average AND SIGNIFICANTLY LOWER MEAN WORKLOADS (emphasis mine). The strong associations between the educational composition of hospitals and other hospital characteristics, including workloads, makes clear the need to control for these latter characteristics in estimating the effects of nurse education on patient mortality."
Wow. Two key things from that statement: a direct acknowledgment that this 'study' is an 'estimation' and an acknowledgment that such an 'estimation' only occurred after 'the need' to highly manipulate the data.
In fact, I think it much more likely to argue that such "co-correlations" makes any 'estimated' conclusions IMPOSSIBLE to verify.
11. Study Conclusions.
This is one of the study's least reported conclusions. See if you agree: "Nurses' years of experience were not found to be a significant predictor of mortality or failure to rescue in the full models." Re-read that and UNDERSTAND the implications of what it means.
The authors admit that their "estimations" can only lead to an "implication" that increased education means better nurses. OK. I'll agree with that. But, because the same study 'factored out' experience, I think it is impossible to estimate how even a fraction of experience affects the conclusions of the study.
Indeed, in order to arrive at its conclusion, the authors must first dismiss the 'conventional wisdom' that experience IS a factor, as they did, in the above statement. Without the above assumption, this whole body of work is worthless. If experience factors in, then the key question cannot be tied simply to education, BUT MUST BE TIED TO BOTH QUALITIES.
And so, the authors find themselves in a conundrum, in which they must first dismiss the importance of experience in order to highlight the importance of education. Amazingly enough, their study reached both conclusions: experience is meaningless to patient outcomes and THEREFORE education level is, by itself, a measurable standard.
The problem with that is, once experience is dismissed, the correlation between education, experience, and patient outcomes is NOT part of this study. Even if you COULD credibly claim that there is no correlation between experience and outcomes (a silly claim), once you add education level into the consideration, you create a new dynamic. By dismissing experience from the equation the study also dimisses its own results, which NOW have the effect of ascribing the results and effects of a real-life system (education AND experience vs outcomes) to a completely different and hypothetical system (education alone vs outcomes).
In short, the claim that experience is not a factor and can be excluded from the study of education's impact on quality is the equivalent of stating that nature is not a factor and can be isolated from nurture in the study of human behavior. In truth, the concepts are much too intricately linked for bland assurances of non-bias in the elimination of part of either equation.
Also not taken into consideration is alternative educational pathways, such as non-BSN bach degree'd RNs (To include both Accel Programs and '2nd Career' ADN nurses.)
The study also fails to note that many BSNs are prior ADN students. While the subset of BSN included ADN graduates, the subset of ADN graduates ALMOST NEVER includes BSN graduates. This would obviously skew the data unless this characteristic were isolated from the data set. In fact, the data set isn't a pool of RNs but pt discharge records and there is no way included within this study to make a distinction.
Given the broad range of said experiences and educations within nursing, negating those experiences and educational pathways also serves the purpose of negating the validity of the study itself.
My conclusion:
Saying that education is a bigger factor than experience in ANYTHING is the same as saying that nurture is a bigger factor than nature in ANYTHING. The relationships are so intricately linked as to be inseparable. As a result, these types of arguments rise to the level of philosophy.
This study claims the ability to make such distinctions, using incomplete and highly manipulated (133 ways by its own admission) data and applying that data only to hypothetical situations.
This is not science; it's propaganda.
Simply put, this flawed and un-reproducible study is worthless as anything BUT propaganda. And that's the bottom line.
~faith,
Timothy.