Clinical problem-solving. A little math helps the medicine go down

Determining clinical decision thresholds for HIV-positive patients suspected of having tuberculosis

Clinical decision thresholds may aid the evaluation of diagnostic tests but have rarely been determined for tuberculosis (TB). We presented clinicians with six web-based clinical scenarios, describing patients with HIV and possible TB at various sites and with a range of clinical stability. The probability of disease was varied randomly and clinicians asked to make treatment decisions; threshold curves and therapeutic thresholds were calculated. Test and treatment thresholds were calculated using Bayes theorem and the diagnostic accuracy of Xpert MTB/RIF. We received 165 replies to our survey. Therapeutic thresholds vary depending on the clinical stability and site of suspected disease. For inpatients, it ranges from 3.4% in unstable to 79.6% in stable patients. For TB meningitis, it ranges from 0% in unstable to 51.4% in stable patients and for pulmonary TB in outpatients it ranges from 29.1% in unstable to 74.5% in the stable patients. Test and treatment thresholds vary in a similar way with test thresholds ranging from 0 in unstable patients with suspected meningitis to 8.2% for stable inpatients. Treatment thresholds vary from 0 for unstable patients with suspected meningitis to 97% for stable inpatients. Therapeutic thresholds for TB can be determined by presenting clinicians with patient scenarios with random probabilities of disease and can be used to calculate test and treatment thresholds using Bayes theorem. Thresholds are lower when patients are more clinically unstable and when the implications of inappropriately withholding therapy are more serious. These results can be used to improve use and evaluation of diagnostic tests.

Rationality, practice variation and person-centred health policy: a threshold hypothesis

Variation in practice of medicine is one of the major health policy issues of today. Ultimately, it is related to physicians' decision making. Similar patients with similar likelihood of having disease are often managed by different doctors differently: some doctors may elect to observe the patient, others decide to act based on diagnostic testing and yet others may elect to treat without testing. We explain these differences in practice by differences in disease probability thresholds at which physicians decide to act: contextual social and clinical factors and emotions such as regret affect the threshold by influencing the way doctors integrate objective data related to treatment and testing. However, depending on a theoretical construct each of the physician's behaviour can be considered rational. In fact, we showed that the current regulatory policies lead to predictably low thresholds for most decisions in contemporary practice. As a result, we may expect continuing motivation for overuse of treatment and diagnostic tests. We argue that rationality should take into account both formal principles of rationality and human intuitions about good decisions along the lines of Rawls' 'reflective equilibrium/considered judgment'. In turn, this can help define a threshold model that is empirically testable.

Should malaria treatment be guided by a point of care rapid test? A threshold approach to malaria management in rural Burkina Faso

BACKGROUND

In Burkina Faso, rapid diagnostic tests for malaria have been made recently available. Previously, malaria was managed clinically. This study aims at assessing which is the best management option of a febrile patient in a hyperendemic setting. Three alternatives are: treating presumptively, testing, or refraining from both test and treatment. The test threshold is the tradeoff between refraining and testing, the test-treatment threshold is the tradeoff between testing and treating. Only if the disease probability lies between the two should the test be used.

METHODS AND FINDINGS

Data for this analysis was obtained from previous studies on malaria rapid tests, involving 5220 patients. The thresholds were calculated, based on disease risk, treatment risk and cost, test accuracy and cost. The thresholds were then matched against the disease probability. For a febrile child under 5 in the dry season, the pre-test probability of clinical malaria (3.2%), was just above the test/treatment threshold. In the rainy season, that probability was 63%, largely above the test/treatment threshold. For febrile children >5 years and adults in the dry season, the probability was 1.7%, below the test threshold, while in the rainy season it was higher (25.1%), and situated between the two thresholds (3% and 60.9%), only if costs were not considered. If they were, neither testing nor treating with artemisinin combination treatments (ACT) would be recommended.

CONCLUSIONS

A febrile child under 5 should be treated presumptively. In the dry season, the probability of clinical malaria in adults is so low, that neither testing nor treating with any regimen should be recommended. In the rainy season, if costs are considered, a febrile adult should not be tested, nor treated with ACT, but a possible alternative would be a presumptive treatment with amodiaquine plus sulfadoxine-pyrimethamine. If costs were not considered, testing would be recommended.

Effect of applying a treatment threshold in a population. An example of pulmonary tuberculosis in Rwanda

PURPOSE

Clinicians often think treatment thresholds should be adapted to the setting. We intended to explore the effect in terms of harm because of false negatives and true and false positives of the application of a treatment threshold for pulmonary tuberculosis from a patient's perspective at different prevalence levels in a developing country.

METHODS

In a cohort of 300 patients with chronic cough, we estimated the prevalence of pulmonary tuberculosis, and the sensitivity and specificity of key predictors with latent class analysis (LCA). We computed the post-test probability of individual patients based on these data. With disease- and treatment-related mortality and morbidity, and without cost or regret, we calculated the break-even point of disease probability where treating versus not treating resulted in similar total harm from the patient's perspective. We estimated the total harm of applying this threshold to the cohort, and to hypothetical settings with different disease prevalence.

RESULTS

The threshold was computed at 0.026, suggesting treatment for all patients of the cohort. Hypothetically lowering the prevalence showed that the lowest total harm in the cohort always coincides with this threshold, but that numbers of treated patients drop considerably.

CONCLUSION

For pulmonary tuberculosis a decision threshold solely based on utilities without cost or regret leads to a very low threshold. The lowest total harm is found always at this disease probability, irrespective of the distribution of the patients. Although these findings might suggest an excess prescription at reference level, this is not the case in settings with lower prevalence.

Evaluation of latent class analysis and decision thresholds to guide the diagnosis of pediatric tuberculosis in a Rwandan reference hospital

SETTING

A pediatric ward of a university hospital in Kigali, Rwanda, a region with a high HIV seroprevalence.

OBJECTIVE

To estimate the diagnostic accuracy of symptoms, signs, and paraclinical investigations for tuberculosis in children, and to propose a clinical rule based on the results.

DESIGN

During a 2-year period all children with cough for more than 2 weeks and/or fever for more than 2 weeks and/or reported weight loss were prospectively included. A set of clinical and paraclinical data were analyzed with latent class analysis. Comparison of post-test probability based on this analysis with a therapeutic threshold for TB was used to develop a guideline.

RESULTS

In the 309 children HIV prevalence was 56%, bacteriology was positive in 9%, and the tuberculin skin test (TST) was >10 mm in 20%. TB prevalence was 32%. Bacteriology and TST had a specificity of 97% and cough had a sensitivity of 91%. Decision analysis suggests treating children presenting one of the inclusion criteria, combined with positive bacteriology or TST >10 mm or contact with a TB patient.

CONCLUSIONS

Latent class analysis confirmed earlier identified predictors for TB and allowed development of an easy to use clinical rule, applicable in reference hospitals of countries with high HIV endemicity.

Evaluation of the GIDEON Expert Computer Program for the Diagnosis of Imported Febrile Illnesses

Objective. The authors evaluate the performance of the expert system Global Infectious Diseases and Epidemiology Network (GIDEON) in diagnosing febrile illnesses occurring after a stay in the tropics. Methods. One investigator (E.B.) entered into the program the collected characteristics of 161 febrile travelers randomly extracted from a database of 1842 cases prospectively included during a study on imported fever. Accuracy was considered acceptable if the correct diagnosis appeared in the top 5 GIDEON ranking list. Interuser agreement was assessed by J.V.d.E. and J.M., who also entered the data of the first 50 sample cases with an established diagnosis. Results. The sample was epidemiologically and clinically representative of the whole cohort. An infectious etiology had been established in 129 cases; diagnosis was unknown in 31 cases and non-infectious in 1 case. GIDEON generated a median of 29 diagnoses per case, including 23 with a probability lower than 1%. Accuracy was acceptable in 64% of the 129 fevers with infectious etiology. It tended to decrease when more than 3 findings were entered per case. Eleven (8%) severe conditions were rejected by GIDEON because non-disease-related characteristics had been introduced. In other cases, the posttest probability was inadequately affected by the insufficient weight of absent relevant findings. Interuser agreement was good for acceptable accuracy and final ranking (kappa=0.83 and 0.72, respectively). Conclusion. The performance of GIDEON in diagnosing imported fever is relatively good and reproducible but is impaired by some conceptual weaknesses. Its use might be hazardous for inexperienced physicians.

Bridging the gap between clinical practice and diagnostic clinical epidemiology: pilot experiences with a didactic model based on a logarithmic scale

Rationale From general practitioners to academic staff, clinicians continue to have difficulties in applying clinical epidemiology in their everyday work. They do not fully understand the logical rules behind the numbers and they do not recognize these rules in their work. We present a new model where the pre-test and the post-test probabilities are converted to log10 of odds, and the likelihood ratio (LR) to its own log10. Methods Following Bayes' theorem, adding the log10LR to the log10 pre-test odds gives the log10 post-test odds, which can easily be represented on a logarithmic scale. In addition, by rounding the log10LR to half the unit, we create classes of discriminative power of tests, close to intuitive estimation. This model generates also a user-friendly diagram, adding considerably to the understanding of Bayes' theorem. We evaluated the effect of the rounding, the current use of the classical model and the acceptability of the new model. Results Rounding 10 disease characteristics to half the unit gives an absolute error of less than half a unit. After six explanations of Bayes' theorem, only 6/16 medical students were capable of answering simple questions about predictive value. When asked about weight of disease characteristics, no one of the 50 clinicians mentioned sensitivity, specificity, predictive value or LR. With the new model, more than 80% of trainees found medical decision making easier to understand and recognized the theory in their practice. Conclusions We conclude that our model of diagnostic clinical epidemiology offers a logical environment for an easy and rapid assessment of the evolution of disease probability with consecutive tests, providing a scientific format for 'qualitative' clinical estimations.

Why are clinicians reluctant to treat smear-negative tuberculosis? An inquiry about treatment thresholds in Rwanda

PURPOSE

The diagnosis of tuberculosis remains controversial between clinicians and public health officers. Public health officials fear to treat too many patients; clinicians fear that truly diseased will be denied treatment. We wondered whether an analysis of the treatment threshold could help making the often intuitive decision to treat smear-negative cases more evidence based.

METHODS

Eighteen clinicians and 10 public health specialists were asked for an intuitive estimate of their treatment threshold for tuberculosis and of key determinant factors for this threshold: the magnitude and subjective weight of mortality and morbidity due to both the disease and the treatment and risk and cost of the latter. With these factors, the authors calculated treatment thresholds and compared them to the intuitive thresholds of the interviewees. A prescriptive threshold was calculated based on literature data, omitting cost and subjective factors.

RESULTS

The median overall intuitive treatment threshold was 52.5%, the calculated 11.9%, and the prescriptive 2.7%. For 2 factors, public health officers provided significantly lower values than clinicians: cost of treatment (median = 20 dollars v. 300 dollars; U = 2.5; P = 0.0002); cost of life (median = 500 dollars v. 5000 dollars; U = 17.5; P = 0.009).

CONCLUSION

These results suggest that clinicians and public health officers estimate wrongly the threshold even when using their own subjective estimate of influencing factors. Omitting treatment cost and subjective weight of provoked harm can result in a very low threshold. Sound training in threshold principles and providing tools to correctly assess data might help in making better decisions in tuberculosis in developing countries.