Monitoring in chronic disease: a rational approachBMJ 2005; 330 doi: https://doi.org/10.1136/bmj.330.7492.644 (Published 17 March 2005) Cite this as: BMJ 2005;330:644
- 1 University of Oxford, Department of Primary Health Care, Oxford OX3 7LF UK
- 2 Screening and Test Evaluation Program, School of Public Health, University of Sydney, Camperdown, NSW 2006, Australia
- Correspondence to: P Glasziou
- Accepted 20 December 2004
“Know which abnormality you are going to follow during treatment. Pick something you can measure.”Meador C. A Little Book of Doctors' Rules.Lyons: IARC Press, 1999
The ritual of routine visits for most chronic diseases usually includes monitoring to check on the progress or regress of the disease and the development of complications. Such checks require that we choose what to monitor, when to monitor, and how to adjust treatment. Poor choices in each can lead to poor control, poor use of time, and dangerous adjustments to treatment. For example, an audit of serum digoxin monitoring in a UK teaching hospital more than 20 years ago showed that the logic behind more than 80% of the tests requested could not be established, the timing of tests reflected poor understanding of the clinical pharmacokinetics, and about one result in four was followed by an inappropriate clinical decision.1 Improvements are possible. For example, a computerised reminder of inappropriate testing reduced the volume of testing for the concentration of antiepileptic drugs by 20%2; a decision support system for anticoagulation with warfarin led to an improvement from 45% to 63% of patients being within target range3; and quality control charts for peak flow measurements for people with asthma could detect exacerbations four days earlier than conventional methods.4 Given the extent of monitoring, even modest improvements are likely to improve benefits for patients and may reduce costs.
Monitoring is periodic measurement that guides the management of a chronic or recurrent condition. It can be done by clinicians, patients, or both. In Australia, monitoring comprises between a third and half of all tests ordered in general practice and outpatients (Pirozzo, personal communication, 2002). Despite the considerable staff time and resources involved, monitoring is a surprisingly understudied area. We review the current literature (based on a Medline search using the terms “monitor*“, “therap*“ or “treat*“; “limit*“ or “threshold*“; “chronic” and a check of references of relevant papers found) on the clinical uses of monitoring, and develop some principles for good monitoring strategies.
Should we monitor at all?
Although intuitively monitoring should be beneficial, clinicians accept non-monitoring in many areas. Aspirin, for example, is used to prevent stroke without assessing platelet aggregation. Establishing benefit for patients is important, as it must be balanced against the downsides of monitoring, such as the inconvenience and costs, and the impact of false positive and false negative (monitoring) results that can lead to inappropriate or delayed actions. As with other interventions, this ideally requires a randomised controlled trial. The benefit should preferably be measured by outcomes that are relevant to patients rather than time spent in the target range (a surrogate marker). For example, in a trial of drug monitoring in epilepsy, monitored patients were more often within control limits than unmonitored patients (8% v 25%), but the proportion remaining free of seizures did not change (38% v 41%).5 However, few monitoring practices have undergone such trials.
Monitoring is a complex intervention. It can have an impact through several means, including improvement of adherence, better selection of treatments based on individual response, better titration of treatment, and patients' learning about non-treatment factors that alter the condition's control. Because of this complexity, any trial should be preceded by developmental work to understand the optimal strategies.
Monitoring and adjustment may be controlled by clinicians or patients. For patients, monitoring may provide a signal for action or simply provide motivation to adhere to treatment. For example, a recent trial in which some diabetes patients were randomised to self monitoring of risk factors showed a better achievement of target measurements of blood pressure, low density lipoprotein cholesterol, and HbA1c; and a reduction in clinical events.6 However, a randomised trial of the motivational effect of cholesterol measurement in general practice in the United Kingdom showed only negligible benefit—a difference of 0.1 mmol/l in total cholesterol measurements.7
The optimal process is not straightforward. A recent study compared three modes of peak flow monitoring in childhood asthma: regular daily measurements, measurement only when symptomatic, and monitoring only at times of symptoms. The second group—children randomised to measure their peak expiratory flow rate only when symptomatic—had lower asthma severity scores, fewer days of symptoms, and fewer healthcare visits than children randomised to either daily peak flow monitoring or monitoring only at times of symptoms.8 The study showed that peak flow monitoring is helpful but that the less intensive monitoring regimen is preferable.
The phases of monitoring
The objective and methods of monitoring change over the course of treatment. This course of monitoring can be usefully divided into the five phases listed in table 1.
Figure 1 shows a control chart for these stages. At (a) we first note the abnormal measurement and begin a quick series of measurements before treatment to confirm the abnormal result; (b) then, if appropriate, initiate treatment and monitor at short intervals to check response and achieve control; (c) but once control is achieved, the intervals may be longer (d), although this may be supplemented by patients' self monitoring (small arrows) (e); but when one measurement is more than 3 standard deviations (SD) or two measurements are more than 2 standard deviations from target, we adjust therapy to re-establish control and shorten the re-check interval; and finally (f), if treatment becomes unnecessary, a period of monitoring after its cessation may be required. We now look at these phases in more detail.
Monitoring before treatment should establish the need for treatment and then a baseline to judge the response to treatment or changes in the patient's condition. Treatment should not start until sufficient measurements for a firm baseline have been obtained. This firm baseline confirms that the degree of abnormality is beyond the initiation threshold. Serial measurements often “normalise” before treatment for several reasons, such as training effects (for example, with a peak flow meter), accommodation to measurement (for example, with blood pressure measurement), and, perhaps most importantly, regression to the mean (the tendency of repeat tests to be closer to normal9). For example, in the first biennial check in the Framingham study, the average blood pressure fell by 3.4 mm Hg systolic and 2.4 mm Hg diastolic.10 Among 99 Dutch patients with apparently raised blood pressure, re-measurement resulted in average reductions of 9 mm Hg systolic and 4 mm Hg diastolic.11 The same study showed that two measurements were sufficient to establish the need for treatment in patients who were well above the initiation threshold. For borderline patients, even four measurements resulted in substantial misclassification.
Initial titration: response, control, and safety
After establishing the baseline, we should set a target and start titration to achieve that target. However, achieving the target is just one of several objectives of the initial titration phase, which should include checking the individual's response to treatment, detecting unacceptable adverse effects, and achieving the desired target range.
This initial monitoring checks adequate response to treatment—that is, whether it “works” as expected on the basis of clinical trials in other patients. Sometimes individual response to treatment can be predicted by other measurements, such as genetic testing (for example, the dose requirement on initiating warfarin treatment is titlely related to CYP2C9 gene variants12). Sometimes it can be predicted by pharmacokinetic studies—for example, in initiating tricyclic antidepressants, short term measurement of the drug concentration can characterise individual metabolism and guide the long term dose used.13 However, personalising treatment usually requires some pragmatic trial and error. We can therefore think of this phase as an “n = 1” trial.14 Ideally, the estimate of effect will be based on both the measurements in that patient and the known effect from trials.15
The initiation phase should also detect immediate or short term adverse effects.16 Measures of potential harm should hence be assessed and monitored. Almost half of the medicines in the electronic Medicines Compendium (http://www.medicines.org.uk/) include a suggestion for some monitoring. Many drugs affect renal function, and regular monitoring of creatinine and electrolytes is widely recommended. A more specific example is treatment with clozapine, for which white cell counts are measured on a weekly basis for the first few months to detect agranulocytosis, which occurs in 0.8% of patients. However, the rationale for the timing of measurement is seldom explicit. The criteria for monitoring for adverse effects are similar to those for screening, including that the effect is serious, that a simple test is available, that earlier detection is predictive, and that change in treatment leads to a better outcome.17
Monitoring during treatment
Once the target is achieved, the objective of monitoring is to ensure that measurements stay within reasonable limits, called control limits. The control limits ensure that we detect real changes in the level of the target measure while minimising false positives resulting from variable measurements in the short term or errors in the technical measurement. The degree of short term variation can be estimated from population studies, subpopulations, or from the individual's own measurements. Because extreme measurements are unlikely to be due to variability of short term measurements, they may justify action to re-establish control. One approach suggested by statistical control theory is to consider that a shift from control has occurred if a single measurement is outside an upper and lower control limit of 3 standard deviations, or if two or three successive measurements are more than 2 standard deviations from the target.18 Figure 1 shows these two sets of action thresholds; one for action (SD 3) and one for re-measurement (SD 2), with action if the repeat result is also more than 2 standard deviations from target.
Monitoring during treatment can be less frequent than during the initiation phase. The interval depends on the probability of being outside the control limits, which in turn depends on both random drift and systematic changes (progression or regression of disease).
The measurements intervals may be shorter than the decision interval. For example, monitoring measurements of blood pressure might be done daily by patients, but the decision made at a monthly consultation with the doctor. This is illustrated in figure 1, where multiple measures (3(b)—small arrows) occur between the decision points (3(a)—large arrows). Ideally a graphical presentation, such as a control chart, should be used to aid recording and deciding treatment changes.
Adjustment to re-establish control
When a clear drift beyond the control limits occurs, we should re-establish control. As in the initiation phase, a shorter measurement interval is generally warranted until control is re-established.
We have already cited several examples (for example, monitoring digoxin therapy) in which audits have shown that clinical decisions taken as a result of monitoring are suboptimal—sometimes tending to the conservative. A New Zealand study of digoxin monitoring showed that 53% of ordered measurements were inappropriately timed and that 5% of the ordered measurements led to inappropriate dose adjustments.19 Patients too find this difficult: a survey of diabetes educators showed that correct adjustment of insulin dosage is the single hardest skill to teach.20 This may be because of the lack of a planned and explicit response to test results or because the tests are seldom considered in context—they usually arrive as a single value, without reference to past values or a clear and useful statement about the variability of measurements. The appropriate degree of adjustment may be helped by nomograms or computer algorithms.
Cessation of treatment
Most therapies are not lifelong. Monitoring should inform stopping decisions. Again, such decisions are usually based on a threshold level—either a negative outcome (such as an adverse effect on renal function or frequency of epileptic seizures exceeding a threshold), or a positive outcome (such as pain relief) that falls below a minimum threshold. However, the precise thresholds chosen and the monitoring interval depend on the phase of treatment. Cessation of treatment mirrors the first phases of pre-treatment and initiating treatment. A decision to stop based on current risks and control is made, treatment is withdrawn (perhaps in stages), and, after a “washout” period, the patient is rechecked to ensure that treatment does not need to be restarted.
For all phases, several choices have to be made in devising a good monitoring strategy: whether to monitor at all, the choice of measurement(s), the choice of target range, the choice of measurement interval, and who should monitor.
Monitoring the wrong target is obviously a fundamental error. Often various possible monitoring tests will be available, and there are advantages to choosing one main measurement (or a composite measure of several) to guide changes in management (fig 2). The criteria for the choice of measurement include the following.
Is it a good predictor of clinically relevant outcomes? Monitoring measurements are surrogate markers for the patient relevant outcomes.21
Can it detect changes in risk early? Risk predictors will vary in their responsiveness to the beneficial impact of treatment, with a good monitoring measurement providing an early indication of risk change.22
Is the random variability acceptable or can it be made acceptable by repeated measurements?
Is it sufficiently affordable, accessible, and acceptable to patients?
Choosing the monitoring interval
The different phases require different monitoring intervals (table 1). The interval is shortest for the pretreatment phase, longest for the maintenance phase, and intermediate for the other three phases.
For the pretreatment phase, measurements need to be far enough apart to make allowance for the short term variability within and between days. For example, the multiple readings taken when monitoring blood pressure for 24 hours give an accurate picture over a full day but do not capture some patients' considerable variability from day to day. For remeasurement during the treatment phases, we must know the time to therapeutic response (allowing for both pharmacokinetics and pharmacodynamics)—for example, the interval for statins is about six weeks23 and for angiotensin converting enzyme inhibitors for blood pressure about three weeks.24
Who should monitor?
Self monitoring by patients is becoming more common. Such monitoring may be for motivation—for example, by using pedometers to monitor physical activity or home blood glucose measurement for people with non-insulin dependent diabetes25; for providing a clinician with between visits measurements—for example, home blood pressure monitoring26, or home glucose monitoring; or for self adjustment of therapy—for example, peak flow rates to trigger actions in asthma patients27 or home blood glucose for insulin adjustments. As the examples show, monitoring of patients can be done for all three purposes.
Although monitoring is common in clinical practice, the principles of monitoring have not been well conceptualised, which in turn has led to suboptimal care. Chronic care could potentially be improved (and often at reduced costs) if for each chronic disease we determined whether and how monitoring was necessary, set explicit monitoring ranges and provided appropriate graphical representations that aided decision making, recognised the need for different optimal intervals for different phases, and understood better when and how to adjust treatment to avoid the increases in variability caused by overadjustment. Health professionals working in chronic care need to understanding these principles better, and systems needs to be improved, including use of appropriate decision aids that have been shown to improve monitoring care.28
Additional educational resources
For patients and clinicians
National electronic Library for Health (http://www.nelh.nhs.uk/)—contains UK relevant guidelines on monitoring for many areas. A search on “monitoring” and choosing “evidence” provides the evidence base for a number of monitoring topics
The BTS/SIGN British Guideline for Asthma (www.brit-thoracic.org.uk/sign/index.htm)—few long term conditions have appropriate monitoring charts, but asthma is one of the better served. A printable peak flow diary is downloadable from the websites of the British Thoracic Society and Scottish Intercollegiate Guidelines Network (www.brit-thoracic.org.uk/sign/mainframe_download.html)
Westgard QC (http://www.westgard.com/)—a website for laboratory tests, which includes a database of within person variation for over 100 tests (see “Biologic Variation Database”), and the Westgard rules for detect readings beyond control limits
Monitoring aims to establish the response to treatment, detect the need to adjust treatment, and detect adverse effects
Monitoring is not always necessary or beneficial and can lead to inappropriate changes
Control charts help distinguish natural variability from true change and reduce unnecessary adjustment
Monitoring for both benefit and harm is important, preferably with a single measurement
The interval between measurements varies between phases and is shorter after changes in treatment
We thank Andrew Farmer, Jeffrey Aronson, and Tom Fahey for detailed comments on various drafts, and Susan Jack who help with initial searches and problem formulation.
Contributors PG conceived the study, and all authors contributed to the ideas and writing. PG is guarantor.
Funding PG and LI were in part funded by the Australian National Health and Medical Research Council for this work, as part of a programme grant 211205 on testing.
Competing interests None declared