Paul P. Glasziou, MBBS, PhD; Les Irwig, MBBS, PhD; Stephane Heritier, PhD; R. John Simes, MBBS, MD; Andrew Tonkin, MBBS, MD; LIPID Study Investigators
Acknowledgment: The authors thank Tim James for local cholesterol level testing data; Katy Bell, Jorgen Hilden, Martin Turner, Andrew Hayen, and members of the LIPID management committee—David Sullivan, Harvey White, Paul Nestel, and David Colquhoun—for helpful comments; and Rhana Pike for editorial work.
Potential Financial Conflicts of Interest: None disclosed.
Reproducible Research Statement:Study protocol: The original 2-page proposal is available from Dr. Glasziou (e-mail, firstname.lastname@example.org). Statistical code: Available from Dr. Heritier (e-mail, email@example.com). Data set: Not available.
Requests for Single Reprints: Paul P. Glasziou, MBBS, PhD, Centre for Evidence-Based Medicine, Department of Primary Health Care, University of Oxford, Old Road Campus, Oxford OX3 7LF, United Kingdom; e-mail, firstname.lastname@example.org.
Current Author Addresses: Dr. Glasziou: Centre for Evidence-Based Medicine, Department of Primary Health Care, University of Oxford, Old Road Campus, Oxford OX3 7LF, United Kingdom.
Dr. Irwig: Screening and Test Evaluation Program, School of Public Health, University of Sydney, Sydney, Australia.
Drs. Heritier and Simes: NHMRC Clinical Trials Centre, University of Sydney, Sydney, Australia.
Dr. Tonkin: Department of Epidemiology & Preventive Medicine, Monash University, Melbourne, Australia.
Author Contributions: Conception and design: P.P. Glasziou, L. Irwig.
Analysis and interpretation of the data: P.P. Glasziou, L. Irwig, S. Heritier, R.J. Simes, A. Tonkin.
Drafting of the article: P.P. Glasziou, L. Irwig, S. Heritier.
Critical revision of the article for important intellectual content: P.P. Glasziou, L. Irwig, R.J. Simes, A. Tonkin.
Final approval of the article: P.P. Glasziou, L. Irwig, S. Heritier, R.J. Simes, A. Tonkin.
Statistical expertise: P.P. Glasziou, S. Heritier.
Obtaining of funding: P.P. Glasziou, A. Tonkin.
Administrative, technical, or logistic support: R.J. Simes.
Collection and assembly of data: P.P. Glasziou.
Glasziou P., Irwig L., Heritier S., Simes R., Tonkin A., ; Monitoring Cholesterol Levels: Measurement Error or True Change?. Ann Intern Med. 2008;148:656-661. doi: 10.7326/0003-4819-148-9-200805060-00005
Download citation file:
Published: Ann Intern Med. 2008;148(9):656-661.
Cholesterol level monitoring is a common clinical activity. Because indications for treatment have been widening over the past decade, cholesterol-lowering medications have become some of the most widely used and expensive pharmaceutical items, and cholesterol screening, treatment, and monitoring have increased. For example, lipid panels were the third highest contributors to Medicare testing growth between 2000 and 2004, with a 61% increase in volume and a 65% increase in cost (1). Previous studies have suggested that, because of measurement error, frequent monitoring is just as likely to mislead when trying to decide whether changes in treatment are needed (2).
Institute of Digestive Disease, Xijing Hospital, Fourth Military Medical University, XiÂ¡Â¯an, China
May 9, 2008
Questions to authors
To the editor:
As Glasziou et al report, after the initial decrease in cholesterol level in response to treatment, subsequent cholesterol level monitoring may be much less frequent than is currently recommended. They show that much of current testing will detect only false-positive results, which are related to either short-term biological variation or analytic error . However, they don't clearly show the factors inducing false-positive results and the way to decrease variation.
We would like to address some questions. Firstly, could the authors comment on the statistical factors, which induce the differences observed? In our opinions, the variation in cholesterol levels may result from other factors except for pravastatin treatment, such as other medication, physical activity, eating habit, smoking, depression, etc . For example, the patients may receive other medication, which may affect cholesterol levels or enhance (or weaken) the effect of pravastatin. Secondly, we would like to ask the authors whether they try to avoid controllable biological variation. For example, the patients should not drink alcoholic beverages when detecting serial cholesterol concentrations. Thirdly, could the authors comment on the methods to decrese false-positive results? Lastly, this study is based on the participants from Australia and New Zealand, so the results may not be applicable for Asians. Could the authors comment on the effect of racial diversify on the variation in cholesterol levels?
1. Glasziou PP, Irwig L, Heritier S, Simes RJ, Tonkin A; LIPID Study Investigators. Monitoring cholesterol levels: measurement error or true change? Ann Intern Med. 2008 May 6;148(9):656-61.
2 Bogers RP, Bemelmans WJ, Hoogenveen RT, Boshuizen HC, Woodward M, Knekt P, et al. Association of overweight with increased risk of coronary heart disease partly independent of blood pressure and cholesterol levels: a meta-analysis of 21 cohort studies including more than 300 000 persons. Arch Intern Med. 2007 Sep 10;167(16):1720-8.
Medwin Hospital, Hyderabad, India-500001
May 18, 2008
Monitoring Lipid lowering treatment
This is an interesting study showing the concept of signal:noise ratio in cholesterol estimation and the need to increase the time to follow up lipid estimation. Present day guidelines of more frequent monitoring may be better suited to clinical practice settings. Effective cholesterol lowering is an important part of lowering macrovascular disease risk. Periodic monitoring with plasma lipid profile can document the efficacy of lipid lowering and indicate need to increase dose or supplement other drugs. Prolonging the time gap between estimations more than one year can lead to long time periods when patient is subjected to elevated lipid levels and its detrimental effects. Drug and dietary compliance is an important issue in lipid lowering therapy which is often life long. Periodic lipid level estimation can help in reinforcing these concepts and to motivate patients to be compliant. Long gap between serial monitoring may lead to losing follow up and loss of interest in patients regarding their therapy. Although statistical methods do suggest the futility of frequent lipid monitoring in trial settings where patient compliance is likely to be better, day-to-day clinical practice requires methods to ensure patient compliance and regular monitoring at yearly intervals would be a more prudent approach in this regard.
William E Cayley
University of Wisconsin School of Medicine and Public Health
May 23, 2008
Treat based on medication dose not titration to a goal
The finding that "noise" random variations during serial cholesterol level monitoring points to the importance of what exactly has been tested in cholesterol-lowering studies. While guidelines and clinical practice often focus on titrating lipid treatment to obtain specified goal levels (1), the intervention that has actually been tested in many of the important "statin" trials is the administration of a fixed medium or high dose of cholesterol-lowering medication. For example, the WOSCOPS and LIPID studies compared 40 mg pravastatin daily with placebo (2) (3), while PROVE IT and REVERSAL compared fixed high-dose atorvastatin with fixed medium-dose pravastatin (4) (5). Thus, the treatment that has been evaluated in each case is use of a fixed dose of a cholesterol lowering medication, not titration to a specified goal. Evidence-based cholesterol treatment, then, should focus on providing patients with an appropriate statin dose, based on trial data, rather than on a less-studied dose- titration strategy.
(1) Grundy SM, Cleeman JI, Merz CN, Brewer HB Jr, Clark LT, Hunninghake DB, Pasternak RC, et al. Implications of recent clinical trials for the National Cholesterol Education Program Adult Treatment Panel III guidelines. Circulation. 2004 Jul 13;110(2):227-39.
(2) Influence of pravastatin and plasma lipids on clinical events in the West of Scotland Coronary Prevention Study (WOSCOPS). Circulation. 1998 Apr 21;97(15):1440-5.
(3) The Long-Term Intervention with Pravastatin in Ischaemic Disease (LIPID) Study Group. Prevention of cardiovascular events and death with pravastatin in patients with coronary heart disease and a broad range of initial cholesterol levels. N Engl J Med. 1998;339:1349-57.
(4) Ridker PM, Morrow DA, Rose LM, Rifai N, Cannon CP, Braunwald E. Relative efficacy of atorvastatin 80 mg and pravastatin 40 mg in achieving the dual goals of low-density lipoprotein cholesterol <70 mg/dl and C- reactive protein <2 mg/l: an analysis of the PROVE-IT TIMI-22 trial. J Am Coll Cardiol. 2005 May 17;45(10):1644-8.
(5) Nissen SE, Tuzcu EM, Schoenhagen P, Crowe T, Sasiela WJ, Tsai J, Orazem J, Magorien RD, O'Shaughnessy C, Ganz P; Reversal of Atherosclerosis with Aggressive Lipid Lowering (REVERSAL) Investigators. Statin therapy, LDL cholesterol, C-reactive protein, and coronary artery disease. N Engl J Med. 2005 Jan 6;352(1):29-38.
Justin W. Timbie
HSR&D Center for Clinical Management Research, VA Ann Arbor Healthcare System
June 5, 2008
Incorrect variance estimates
I read with great interest this article as it provided an important variance parameter for a simulation study I am working on. The authors appear to have made two errors, and while neither alters the conclusions of their study, researchers using the reported estimates for other work might come to erroneous conclusions depending on the particular estimates selected.
First, the variance estimates reported on page 659, column 2, paragraphs 3 and 4 are in the wrong units. The units should be mmol^2/L^2 or mg^2/dL^2 (not mmol/L and mg/dL).
Second, the variance estimates given on page 658, column 2, paragraph 4 are incorrect. To convert a variance estimate from mmol^2/L^2 to mg^2/dL^2 one must take the square of the conversion factor since variance is not a linear operator :
Var(cX) = c^2*Var(X).
Thus, the variation in initial response to treatment is not 21.8 mg^2/dL^2 as reported, but:
[(1 mg/dL /.02586 mmol/L)^2]* 0.56 mmol^2/L^2
= 837 mg^2/dL^2
The same error applies to the variance estimates reported on page 659, column 2, paragraphs 3 and 4. The authors do report the correct estimates for the STANDARD DEVIATION of LDL changes throughout the paper, and this is likely due to the fact that the authors made the conversion from mmol/L to mg/dL on the scale of standard deviations (and not variance) where squaring the conversion factor is not needed. Thus, the standard deviation estimates are correct in both sets of units, but only the variance estimates reported in mmol^2/L^2 are correct.
Paul P Glasziou
University of Oxford
July 8, 2008
We thank Justin Timbie for the 2 corrections: all our calculations and writing were done in mmol/l but in the revision we added the US units of mg/dL and did this incorrectly for variances (though standard deviations and means are correct).
Dr. Cayley points out that most of the statin trials have used a fixed dose (4S is an exception) rather than monitoring based titration or adjustment. As our work demonstrates, trials with fixed dose allow the opportunity to assess how cholesterol values increase over time. In those patients who truly have an important increase, a change in treatment can then be considered. Inferences can then be made about monitoring based titration or adjustment, but we do not believe practice must exactly echo the trials, which are designed to maximise power not for optimal practice.
Like Dr. Hong we think it would be desirable to understand the factors that could explain and reduce the variability. Some of the variation is irreducible such as the analytic variation from the laboratory, which has a coefficient of variation (CV) of 2.7% compared to the within-person CV of 7.8%. Only some of the short-term biological variation is explained by the other factors Dr. Hong mentions.
Unfortunately we generally do not know these factors. In the LIPID trial analysis we minimised this variability by both the design (run-in periods and fixed dose therapy) and analysis methods that accounted for patients changing therapy. In clinical practice the variability may be greater. Hence our results may not hold if a patient changes medication or substantially changes diet.
1. Choudhury N, Wall PM, Truswell AS. Effect of time between measurements on within-subject variability for total cholesterol and high- density lipoprotein cholesterol in women. Clin Chem. 1994 May;40(5):710-5.
Brian R Stanley
July 29, 2008
Reducing Measurement Error
Perhaps physicians should take a page from the electrical engineers, who, when confronted with low signal-to-noise ratios, employ a technique known as signal averaging to improve signal detection. In monitoring cholesterol levels in an individual patient, rather than accepting a single measurement, one can use the average of several serial measurements to increase accuracy. The error in estimating the patient's "true" cholesterol decreases roughly as the square root of the number of measurements averaged, if one assumes the patient to be in the same state for each measurement and that the noise is uncorrelated. Albeit informally, clinicians routinely employ this technique when evaluating mesurements of blood pressure, blood sugars, and prothrombin times in anticoagulated patients. Conflict of Interest:
to gain full access to the content and tools.
Learn more about subscription options.
Register Now for a free account.
Cardiology, Dyslipidemia, Coronary Risk Factors.
Results provided by:
Copyright © 2016 American College of Physicians. All Rights Reserved.
Print ISSN: 0003-4819 | Online ISSN: 1539-3704
Conditions of Use
This PDF is available to Subscribers Only