Physician age and outcomes in elderly patients in hospital in the US: observational study
BMJ 2017; 357 doi: https://doi.org/10.1136/bmj.j1797 (Published 16 May 2017) Cite this as: BMJ 2017;357:j1797
All rapid responses
With all due respect, I´m very surprised about the rapid response of Y Tsugawa concerning his publication [1]. In my opinion, the author don´t meet the point of the questions of many rapid responses, his answers are insufficient. For example, the author stated in the first answer to Sharma that “younger hospitalists may be more likely to consult with specialists who deliver better care, and this is part of why patients of younger hospitalists fare better”. My opinion is that older hospitalists provide higher-quality care themselves or consult high-performing specialists. His comments to other rapid responses are inconclusive, too. Gatrell is concerned about “sensational headlines” and stated “not a valid take-away message”. One statement of Tsugawa et al. is: “The lack of association between physician age and patient mortality among physicians with higher volume of patients supports this age related hypothesis.” On the contrary, that lack of association raises major doubt about the study.
Another important point is, that the article [1] looks like a duplicate from the same data sources [2,3] (Maslove, Sakurai). It´s the third article using the same data sources and methodology. Are there any conflicts with GCP?
There are more critical rapid responses about statistical methods (Heston “oldest statistical trick in the book”), about ethical “The whole concept of this study is ageist and repugnant. You might as well design the next one based on the race of the physician” (Gabriella Good) and political questions “There are intense political agendas at work here that the editors should be aware of, lest they publish "politicized" science.” “Academic leaders, particularly those at the various Harvard institutions that were responsible for these articles, know that younger physicians are deeply in debt, more likely to be employed by large hospital systems” (Mittler). “Numerous weaknesses in the methodology employed in these studies were pointed out in peer review. All of these are addressed in rebuttal arguments, but few of them actually resolved.” (Maslove)
JP Newhouse, as one of the coauthors, declared being “ a director of Aetna”. Is Aetna a health care company, which sells health care insurance plans and related services, such as medical and pharmaceutical health?
1) Tsugawa Y, Newhouse JP, Zaslavsky AM, et al. Physician age and outcomes in elderly patients in hospital in the US: observational study. BMJ 2017;357:j1797. doi: 10.1136/bmj.j1797
2) Tsugawa Y, Jena AB, Orav EJ, Jha AK. Quality of care delivered by general internists in US hospitals who graduated from foreign versus US medical schools: observational study. BMJ. 2017 Feb 2;356:j273.
3) Tsugawa Y, Jena AB, Figueroa JF, Orav EJ, Blumenthal DM, Jha AK. Comparison of Hospital Mortality and Readmission Rates for Medicare Patients Treated by Male vs Female Physicians. JAMA Intern Med. 2017 Feb 1;177(2):206-213.
Competing interests: No competing interests
I did not see specific patient age statistics vs physician age groupings. Wouldn't older patients, whose risk of dying soon was higher, want to see their own older doctors? Lots of uncontrolled variables in this study... I also agree with one of the other comments that a patient who knew the end of their life was near would seek care from an older physician that would tend to be more empathetic with a patient of their own age.
Competing interests: No competing interests
We are grateful to Sharma for comments on our recent article on the association between physicians’ age and patient outcomes in the U.S.(1) Although we agree that hospitalists may rely on specialists to treat their patients, the attending physician for a patient is ultimately responsible for the care and outcomes of that patient. Moreover, differences in specialist use do not bias our estimates because hospitalists’ decision to consult a specialist is part of the causal pathway. In other words, compared to older hospitalists, younger hospitalists may be more likely to consult with specialists who deliver better care, and this is part of why patients of younger hospitalists fare better. In addition, from a patients’ perspective, it does not matter whether mortality is lower because younger hospitalists provide higher-quality care themselves or consult high-performing specialists. We were aware that our findings can potentially be confounded by do-not-resuscitate (DNR) directives of patients, but our findings remained unaffected after excluding patients with cancer and those discharged to a hospice, both of which are important predictors of DNR directives. Also, our quasi-randomization approach should address differences in DNR directives across physicians of varying age. The results were also consistent when we studied Medicare beneficiaries aged 64 or younger, supporting the generalizability of our findings.
Li and Chen raised interesting points about additional adjustment variables. We agree seasonality may exist for some conditions. However, it is unlikely that older doctors treat patients during winter (when patients are sicker) and younger doctors practice during other seasons (when patients are less severe), and therefore, this does not introduce bias. For all our analyses, we adjusted for 27 comorbidities (Elixhauser comorbidity index) including renal failure. We did not adjust for right ventricular failure, BMI, and smoking, as information was not available in Medicare claim data. However, we are not aware of any evidence suggesting that older doctors are more likely to treat patients with these conditions, and our quasi-randomization approach demonstrated balance across a wide range of patient characteristics, raising the question of why right ventricular failure, BMI, or smoking should differ.
As Hsu et al. pointed out, our study did not directly evaluate the skills of older hospitalists. However, a previous systematic review has shown that medical skills and knowledge may decline as physicians age.(2) We agree that older physicians may be engaged in management work, but from patients’ point-of-view, it is irrelevant why mortality is higher among older hospitalists (i.e., is it because of depreciation of knowledge or due to part-time clinical work secondary to non-clinical, management responsibilities). The patient mortality in our data may be higher than in previously reported studies because we restricted our sample to non-elective admissions. This was important in order to minimize the impact the bias due to hospitalists selecting their patients (i.e., selection bias).
We agree with Mittler that multiple hospitalists may be involved in each admission, and that is exactly why we examined three different ways of attributing patient outcomes to physicians. We confirmed that our findings are not sensitive to how we attribute patients to physicians. Although we could not demonstrate a causal relationship between physicians’ age and patient mortality, our study was built upon a previous systematic review that found older physicians are less likely to deliver standard care and may not have up-to-date skills and knowledge.(2) We have no intention to politicize science, but we believe that patients have every right to understand physician characteristics associated with high-quality care. Furthermore, there is an active interest in the medical community to better understand how care quality changes over a physician’s career and to identify what interventions are needed to maintain skill.
Liepmann commented that the number of patients each hospitalist treated seemed too small. This is due to our method of attributing patient outcomes to a single doctor for a given hospitalization. Using this approach, only one doctor can be responsible for each admission. For example, if three high-volume hospitalists are working as a team, each hospitalist is treating 603 admissions (201*3=603) either as a primary (i.e., responsible for patient outcomes) or secondary (not responsible for patient outcomes) hospitalist.
Heston asserts that standard deviation (SD) may be more informative than standard errors (SE). We used SE because we were interested in whether patient mortality differs between younger and older doctors, rather than how variable patient mortality was within age category. It is important to note that a large sample size does not introduce biases in any way. Instead, a large sample size allows us to make precise estimates so that even a small difference in patient mortality between younger and older doctors can be detected. We agree that, given precise estimates, it is extremely important to appropriately account for confounders when we use big data. However, given that the large sample size leads to precise estimates without introducing any biases, there is no downside of using the data with a large sample size.
While we agree with Maslove that there is no single rule that can perfectly attribute the quality of care, the methods we used have been studied (3) and used in previous studies.(4) We have not randomly chosen physician characteristics that may be associated with patient outcomes. Instead, we rely on previous scientific evidence that suggested that older doctors may deliver lower-quality care than younger colleagues.(2) We are only interested in physician characteristics that have a plausible causal association with patient outcomes.
Sakurai raised an interesting point. In our study, physicians’ age was the exposure (treatment) variable of interest, whereas physicians’ sex was adjusted for as a confounder, and it is important to distinguish exposure variables and confounders. Confounders are the factors that influence the relationship between exposure and outcome variables, and therefore, regression coefficients of confounders are not meaningfully interpretable. The research questions are also completely different between physician age and sex, each of which was built on previous studies that led to different hypotheses.(1, 5)
References:
1. Tsugawa Y, Newhouse JP, Zaslavsky AM, et al. Physician age and outcomes in elderly patients in hospital in the US: observational study. BMJ 2017;357:j1797. doi: 10.1136/bmj.j1797
2. Choudhry NK, Fletcher RH, Soumerai SB. Systematic review: the relationship between clinical experience and quality of health care. Ann Intern Med 2005;142(4):260-73.
3. Adams JL, Mehrotra A, Thomas JW, et al. Physician cost profiling--reliability and risk of misclassification. N Engl J Med 2010;362(11):1014-21. doi: 10.1056/NEJMsa0906323 [published Online First: 2010/03/20]
4. McWilliams JM, Landon BE, Chernew ME, et al. Changes in Patients' Experiences in Medicare Accountable Care Organizations. N Engl J Med 2014;371(18):1715-24.
5. Tsugawa Y, Jena AB, Figueroa JF, et al. Comparison of Hospital Mortality and Readmission Rates for Medicare Patients Treated by Male vs Female Physicians. JAMA Intern Med 2017;177(2):206-13. doi: 10.1001/jamainternmed.2016.7875
Competing interests: No competing interests
“Physician age and outcomes in elderly patients in hospital in the US: observational study” by Tsugawa et al (BMJ 2017;357:j1797) has spawned sensational headlines about “higher mortality in patients treated by older physicians.” That is not a valid take-away message. As the authors suggest in their introduction, both currency of training and experience contribute to quality patient care.
The authors’ own statements call their conclusion into question: “Our findings might just as likely reflect cohort effects rather than declining clinical performance associated with greater age….”
They acknowledge that among hospitalists caring for high volumes of patients, there was no higher mortality among older physicians. Astonishingly, they then say, “The lack of association between physician age and patient mortality among physicians with higher volume of patients supports this age related hypothesis.” On the contrary, that lack of association raises major doubt about the hypothesis.
If age were the key variable:
• there would also be higher mortality among older physicians with higher volume of patients.
• it would also affect other quality metrics, such as readmission rates. The authors found no significant difference.
• older physicians would have higher mortalities among all patients. The authors reported no mortality difference related to physician age among pneumonia admissions.
A more valid conclusion is that there is increased mortality among non-high volume hospitalists; i.e., decreased mortality among high volume hospitalists. How many of the older physicians also had variables related to volume, such as cutting back hours or working fewer nights?
Will the authors now look at the data to compare other physician characteristics and variables? MD vs DO? Public vs private medical school? Black vs white? Asian vs Hispanic? Male vs female? Gay vs straight? Married vs single vs non-marital relationship? Right-handed vs left-handed? Level of indebtedness vs investment portfolio? So many possibilities for further publications using the same study!
The authors also reported a significant difference in Part B costs among older physicians. Total costs for all admissions in their survey were $800,050,156. If the physicians over age 60 had the same costs per admission as the physicians under age 40, total costs would have been $797,967,880, a savings of 0.26% -- hardly worth mentioning anywhere, much less publishing in The British Medical Journal.
Competing interests: No competing interests
To the Editor:
With great interest, I read the article by Tsugawa et al. To my eyes, however, the current study looks like almost a duplicate from practically the same data sources. The authors had previously reported several studies in almost the same situation.[1,2] In those studies, the authors had reported 30-day mortality with adjustment for physician characteristics, including age. If there was no significant relationship between 30-day mortality and physician age in the previous studies by multiple logistic regression, even in the event of twice the number of patients, the results of the current study may be inconclusive. On the other hand, if a significant relationship had previously been shown, the authors should have reported physician age as a confounder in the previous studies, not in the current study, since these were observational studies.
In addition, the authors should report the influence of physician sex and indicator variables for medical school from which a physician graduated on 30-day mortality, since these confounders were adjusted in the current observational study. It would be of great help for patients to choose physicians if the authors would let them know whether the effects of physician sex and indicator variables for medical school on 30-day mortality were consistent with previous studies.
References:
1. Tsugawa Y, Jena AB, Orav EJ, Jha AK. Quality of care delivered by general internists in US hospitals who graduated from foreign versus US medical schools: observational study. BMJ. 2017 Feb 2;356:j273.
2. Tsugawa Y, Jena AB, Figueroa JF, Orav EJ, Blumenthal DM, Jha AK. Comparison of Hospital Mortality and Readmission Rates for Medicare Patients Treated by Male vs Female Physicians. JAMA Intern Med. 2017 Feb 1;177(2):206-213.
Competing interests: No competing interests
Dr Liepman's is one of several very interesting and critical responses to this study and his point about volume and smaller hospitals well taken
His letter further illustrates the perils of extrapolating from one health system to another, however. From an NHS secondary care doctor perspective, a geriatrician or acute internal medicine physician or general internal medicine physician in full time practice would routinely look after multiples of the 200 patients a year figure a year. It is very hard to work out how anyone could be gainfully employed seeing volumes less than 200 and in turn just highlights how very different health systems can be
David Oliver
Competing interests: No competing interests
This is Tsugawa et al's third article using the same data sources and methodology to demonstrate an association between physician characteristics (sex, age, and location of training), and patient outcomes. Thus far, every hypothesis the authors have tested has been found to be statistically significant, and therefore presumed to be important. These results speak more to the size of the dataset being examined, than to the validity of the hypotheses themselves. If the clinical outcomes studied are associated with 3 out 3 of the factors tested thus far, then we must also ask if they are associated with physician eye colour, or favourite food.
Numerous weaknesses in the methodology employed in these studies were pointed out in peer review. All of these are addressed in rebuttal arguments, but few of them actually resolved. Perhaps the most troubling of these is the method by which the care delivered by a team of physicians was attributed to one single doctor. While three different attribution methods were tested, none of these was the most obvious choice; look only at a representative subset of admissions for which a single physician made claims. While this is likely to be a minority of cases, the original dataset includes 1.2 million patients; even a fraction of these may still constitute a large dataset. The method of care attribution is the lynchpin for all of these studies – getting this part right is essential.
Overall, the recent studies by Tsugawa et al illustrate 2 important points. First, the retrospective analysis of very large datasets is likely to generate results that would be considered statistically significant based on conventional tests. We must look instead to the magnitude of these effects, which in all 3 of these studies is modest at best. Second, these results can easily be misinterpreted, not only by the lay public, but by the medical community as well. These papers are not observational studies per se, but rather retrospective analyses of large datasets collected under non-standardized conditions, for reasons unrelated to the study question. At best, this methodology may point to associations between physician characteristics and patient outcomes, but says nothing about cause and effect. The nuance in this important distinction may indeed have been lost on the 350 media outlets that have reported on these three papers, leaving the general public with the disquieting feeling that their older, male, American-trained hospitalist – whom they had no role in selecting – is putting them at undue risk.
Competing interests: No competing interests
The authors of this article use the oldest statistical trick in the book: using standard errors with a large population in order to come up with a p-value of < 0.05. What is more relevant are the standard deviations. While standard errors may reveal insignificant but interesting large population results, ultimately the data will be applied to individual physicians. To apply the data to individual physicians, the use of standard deviations must be utilized. The authors did not provide the standard deviations, however, based upon their large sample size it appears the standard deviations are very broad. Large standard deviations would mean that there is a huge overlap between physicians of different ages when looking at their clinical outcomes. In fact, the huge overlap would greatly outweigh any age differences, and in effect make age an irrelevant factor. For example, when looking at large populations, a researcher may find a statistically significant difference in math scores, or IQ scores, or empathy scores between men and women or between various races. In these cases, we recognize that individual characteristics are what matters, not the large population differences. Therefore, if someone applied the population differences found to individuals, they would rightly be called racist or sexist. In a similar manner, because individual physician characteristics are likely to greatly outweigh any differences found in these large population studies, anyone applying this data to individual physicians would be inappropriate and unreasonable.
Competing interests: No competing interests
Interesting. "Older hospitalists with patient volume under 200, and especially under 90 admissions/yr have higher death rates." Now, this is odd, because hospitals tend to hire 'just enough' hospitalists to keep them busy. As the authors note, hospitalists tend to work 7/7, about 160 days/year. Where would we find hospitalists with <90 admissions/yr? One answer- small rural hospitals with low inpatient volumes. Small rural hospitals in the US are in trouble. They get paid less for the same care, have trouble finding staff, and their patients are older, sicker, and poorer- a group that tends to do worse in the hospital, despite corrections for income, education, etc.
Another possible explanation for so few admissions is part-time locum physicians, in hospitals unable to attract full time hospitalists. I can easily imagine struggling or poorly run hospitals patching together a hospitalist service entirely of part-timers, who would be expected to have poorer coordination with other phycicians, hospital services, and be less aware of local hospital factors, such as "Don't consult with THAT doctor," or "Be careful on 2East-they have staffing problems." Bad management has pervasive effects.
Indeed, low volume hospitalists at all ages have death rates 20% higher than high volume hospitalists, an effect that increases with age.
So this study could have been titled, "Older part-time hospitalists may work in less-well run hospitals that have fewer resources, are struggling to survive themselves, and despite that have only slightly worse patient outcomes." US health care is broken. Authors can generate articles by finding incidental correlations with poorly run or disadvantaged hospitals.
Competing interests: No competing interests
Re: Physician age and outcomes in elderly patients in hospital in the US: observational study
“You see, but you don’t observe,” Holmes says to Watson in “A Scandal in Bohemia”. It is here that Homes describes the influence of facts on the truth. In a simple, clear pedagogy for which is a defining characteristic of Mr Holmes, each fact is linked to another to create the framework of truth. If these facts are misplaced they are still facts - however, they are just not a true picture of event, they become a false positive. The essence of the case is lost.
Now when I read something in prestigious journals, an observation that I don’t perceive, I wonder, what part of the plot am I missing. Is it simply my particular patient populations that gives me a sample bias? Is it my own personal bigotries that prevent me from seeing, blinders willfully but unknowingly placed upon my eyes to avoid a certain reality? Maybe it is simply sample size, I am not laying my hands on enough patients or strolling the hospital wards long enough to participate in the observations at hand. Or was there an agenda behind the article at hand?
The problem with searching databases to produce a model begins with understanding your own biases, the ones you bring at the start of the project. What are the objectives: to bring forth light, or just use the same old formula to publish. This idea was aptly put forward by Lewis “We simply found ourselves in contact with a certain current of ideas and plunged into it because it seemed modern and successful… writing the kind of essays that got good marks … that won applause. “ The Great Divorce.
As medicine looks to “Big Data” for answers it will find ready made results as so much of it is pre-manufactured prior to being reported. Made to fit by hospital administration under the threat of CMS, it is now there just waiting to be processed by the data miners. But this is not necessarily truth. Side by side with Arthur Conan Doyle, time might be better spent learning from fiction how to read facts and discern truth.
Competing interests: No competing interests