RSS Feeds
-
Lifelong treatment for one millimeter of mercury?
Submit responseThe authors are correct in stating that it is time for us to move beyond binary thinking in the diagnosis of diseases that exist on a continuum. While the authors do not explore the hazards of binary thinking in depth, they do mention the possibility of "overdiagnosis", and cite prostate cancer as a case where overdiagnosis may lead to increased morbidity. Prostate cancer is indeed a striking example of a disease whose treatment may in fact cause more morbidity than the disease itself, but examples can also be drawn from the other diseases mentioned by the authors.
Hypertension and diabetes in particular are of interest as they are so often treated in the clinic. Using a binary approach, a young patient with a systolic blood pressure greater than 140 and no other risk factors will be placed on lifelong antihypertensive therapy. Aside from side effects which may include impotence and depression, antihypertensive therapy may carry additional risks. Thiazide diuretics, which are recommended as first-line treatment, may increase the risk of type 2 diabetes (1) as well as renal cell carcinoma (2). Beta-blockers may similarly increase the risk of type 2 diabetes (1). Drugs used for glycemic control may have their own inherent risks as well, as was recently found to the case with rosiglitazone (3). A risk-prediction approach to the treatment of these diseases would reduce the number of individuals on lifelong treatment with drugs whose long-term effects may not yet be fully understood.
References:
1. Elliot WJ, Peyer PM. Incident diabetes in clinical trials of antihypertensive drugs: a network meta-analysis. Lancet. 2007. 369(9557):201-7.
2. Schouten LJ, et al. Hypertension, antihypertensives and mutations in the Von Hippel-Lindau gene in renal cell carcinoma: results from the Netherlands Cohort Study. J Hyptertens. 2005. 23(11): 1997-2004.
3. Nissen SE, Wolski K. Effect of rosiglitazone on the risk of myocardial infarction and death from cardiovascular causes. N Eng J Med. 2007. 356(24):2457-71.
Conflict of Interest:
None declared
-
A response from the authors
Submit responseWarner, Djulbegovic and Patrick each point to practical problems with a prediction approach. Warner concurs with a point we made in our paper, which is that use of discrete categories (disease / no disease) simplifies clinical management and communication. We also agree with his argument that changes in medical education are needed to help physicians understand and communicate the results of risk prediction. A point of disagreement is that physicians’ and patients’ poor understanding of probabilities is a special problem for the risk prediction approach. For example, even if we use binary diagnostic categories, we would still want to inform the patient about their risk "(Mr. Jones, you have hypertension, which means a 20% risk of having a heart attack)." Conversely, we might use prediction models without reference to numbers at all ( “Mr Jones, you are at high risk of a heart attack so I am going to write you a prescription for some pills)."
Djulbegovic argues that whether we use a binary diagnostic category or a risk prediction model, we still have to choose a threshold to treat a patient. This can cause problems when results are close to the threshold. We would agree that there is room for both descriptive and normative research on decision making near decision thresholds. We also agree with Patrick’s point that we currently “live in a binary world”, and enjoyed his amusing description of the numerous ways in which those outside the examination room force a doctor to think in simple binary terms. We are not naive about the practical challenges of implementing a prediction approach. That said, we must make medical progress in the best interests of our patients and hope that outside forces and structures follow along: we would certainly hate to see, say, the military’s need for specific criteria for service disqualification affect the way we practice medicine.
Swerlick makes a distinction between having symptoms or functional impairment and having only a risk factor for a disease. Although we focused on risk factors, we believe binary diagnostic thinking is also often inappropriate for symptomatic disease. For example, many people have symptoms of depression; a choice of a particular cut-point on a spectrum of severity does not create two natural categories of depressed vs. not depressed. A prediction approach would focus on whether treatment would do more good than harm.
Conflict of Interest:
None declared
-
A Response to "Against Diagnosis"
Submit responseVickers and colleagues discuss an interesting premise in their article “Against Diagnosis (1)”: physicians should consider moving away from the practice of diagnosis and towards the incorporation of risk stratification in their day-to-day practice. This raises several practical issues. The first is that the diagnoses that were discussed are all special cases where the diagnosis has been defined by test results. While many common diagnoses may fit into this category, there are also many others that can be suggested but not proven definitively by routine diagnostic tests (pneumonia and pulmonary embolism are two that come quickly to mind). In these cases, diagnosis reflects a sufficiently high level of certainty that a condition is existent to move forward with management. Creating a discrete categorization greatly simplifies management decisions and communication with patients and colleagues, even though the likelihood of the condition is continuous. The more significant practical issue is the implementation of risk stratification tools. As the authors point out in their illustration of the Framingham Risk Score (2), there are many such tools in existence (The Medical Algorithms Project (3) currently documents in excess of 11,000!). This is daunting in and of itself; some of these algorithms are also quite complex. The APACHE-II score (4), for example, has roughly 20 variables that must be entered. Another potential problem arises when the clinician must interpret the output of a risk stratification algorithm; it has been shown that most medicine residents have a deficient fund of knowledge in biostatistics (5). It is also unclear how to effectively convey information about risk to patients. A recent article found that patients have more difficulty interpreting a “1-in-n” statement than a percentage or frequency term (6). Risk stratification holds great potential as medicine begins to move from a diagnostic to a prognostic framework. Substantial changes in the medical education system will be required to enable physicians to know when to apply these tools, how to interpret the results, and how to convey the information to colleagues and patients. References 1. Vickers AJ, Basch E, Kattan MW. Against diagnosis. Ann Intern Med. 2008;149:200-203. [PMID: 18678847] 2. Wilson PW, D'Agostino RB, Levy D, Belanger AM, Silbershatz H, Kannel WB. Prediction of coronary heart disease using risk factor categories. Circulation. 1998;97:1837-1847. [PMID 9603539] 3. The Medical Algorithms Project. Updated January 2008. Accessed at www.medal.org on 14 August 2008. 4. Knaus WA, Draper EA, Wagner DP, Zimmerman JE. APACHE II: a severity of disease classification system. Crit Care Med. 1985;13:818-829. [PMID 3928249] 5. Windish DM, Huot SJ, Green ML. Medicine residents' understanding of the biostatistics and results in the medical literature. JAMA. 2007;298:1010-1022. [PMID 17785646] 6.Cuite CL, Weinstein ND, Emmons K, Colditz G. A test of numeric formats for communicating risk probabilities. Med Decis Making. 2008;28:377-384. [PMID 18480036] Conflict of Interest:None declared
-
Diagnosis; Disease or risk factor?
Submit responseIn their piece "Against Diagnosis, Vickers et. al. identify problems with the use of a binary approach to the diagnosis of disease. They are to be lauded for their attempt to change how we view the act of diagnosis. However, the authors need to make a distinction between having disease and having a risk factor for disease. A disease state implies existence of symptoms and/or functional impairment. Diagnosis in this context is useful in that it may not only predict the development of further symptoms or impairment, but should also be useful in terms of selecting appropriate treatment and predicting response to therapy. Identification of asymptomatic physical or biochemical parameters which place the patient at risk for future impairment or symptoms perhaps can be better viewed not as disease but as a risk factor. Ideally this would tend to push physicians and patients to ask the obvious question. How big is the risk and how do we know? Conflict of Interest:None declared
-
Medical diagnosis and philosophy of vagueness- uncertainty due to borderline cases
Submit responseI have read with the great interest the paper by Vickers et al [1] “Against diagnosis” in which they note the problem of diagnosis due to difficulties in classification of borderline cases: a single individual or very small differences in the quantities of interest may completely change categorization and consequent actions rather dramatically. Unfortunately, the risk prediction alternative recommended by the authors will unlikely help attenuate the problem of the diagnosis of diseases defined on a continuum. In fact, the problem described by the authors is not trivial and has preoccupied philosophers since the ancient times. It is commonly referred as Sorites paradox [from the Greek word soros (=heap)]: if one removes a single grain from the heap sand, would it be still a heap? Yes. However, as we continue removing a grain of sand at the time, eventually we will reach a point that the heap of sand is not a heap any more. Where one draws the line? Is the heap defined by 10 grains, 9 grains etc? The Sorites Paradox has it origin in vagueness, which deals with unknowability of the borderline statements [2] (such in case described by Vickers et al, about inherent uncertainty of distinguishing clinical consequences between the effect of blood pressure of 140 vs. 139 mmHg). Many solutions have been proposed in the philosophical literature how to deal with Sorites Paradox. A partial, and non-exhaustive list includes fuzzy logic and fuzzy set theory approach, supervaluations ,intuitionistic logic, paraconsistent logic, modal logic, possibility theory, rough sets theory, open texture concept etc [3,4]. The success of the proposed solutions, to some extent, varies depending on the context in which Sorites paradox can apply. How effective these solutions can be in medical arena is not known; a little work has been done related to the role of vagueness in medicine. Vickers et al. are applauded to draw attention of the medical audience to the this important question. With this letter I hope to stimulate application of philosophy of vagueness in medicine, which , despite its importance, has been sorely missing. References 1. Vickers AJ, Basch E, Kattan MW. Against diagnosis. Ann Intern Med 2008;149:200-203 2. Williamson, T. Vagueness. 1994; London:Routledge 3. Hyde D. Sorites Paradox. http://plato.stanford.edu/entries/sorites- paradox/ 4. Sorenson R. Vagueness. http://plato.stanford.edu/entries/vagueness/ Conflict of Interest:None declared
-
It's a Binary World
Submit responseVickers and colleagues fail to mention the greatest impediment to implementing prediction: physicians practice in an binary world that is driven by diagnosis. Those who pay for health care profit from risk management and care little for clinical nuance. A diagnosis of hypertension has a direct impact on the premium that a patient will pay for health or life insurance --whether it is benign (401.1) or malignant (401.0). A diagnosis of depression can make disability insurance unobtainable. A diagnosis of asthma can disqualify a patient from employment or from serving in the military irrespective of how well the asthma is controlled. Computers are binary machines. An Electronic Medical Record demands specific diagnoses with fixed end points to be entered into the computer. Computerize algorithms are routinely used to determine "medical necessity" and "quality care". Woe betide the patient who falls outside of the algorithm. I routinely must explain by phone or by letter why a particular patient requires additional time in the hospital or a plan of treatment to a reviewer who complains that my plans fall outside their (proprietary) guidelines for a particular diagnosis. Often I must first comply with a series of conservative (ie cheaper) treatments (meticulously documented) before my treatment plan will be considered. Medication formularies are driven by similar guidelines and often require pharmacy records (no samples please!) to prove that the cheaper medication was tried first. Agencies uncer contract to provide Quality Reviews mine the same data. My outpatient care of one elderly diabetic generates multiple individual computer generated letters several times a year that ask why the patient is not taking an ACE inhibitor (renal failure), a statin (myopathy), as well as why he still needs a PPI (history of GI bleeding and documented GERD) and have I considered screening for osteoporosis (his insurance policy will not cover the cost of additional medication. Predicition modeling is a useful tool for the individual physician working with his individual patient. But outside of the examination room diagnosis rules. Conflict of Interest:None declared