# Revisiting the relationship between baseline risk and risk under treatment

- Hao Wang
^{1}, - Jean-Pierre Boissel
^{2}and - Patrice Nony
^{2}Email author

**6**:1

**DOI: **10.1186/1742-7622-6-1

© Wang et al; licensee BioMed Central Ltd. 2009

**Received: **07 August 2008

**Accepted: **17 February 2009

**Published: **17 February 2009

## Abstract

### Background

In medical practice, it is generally accepted that the 'effect model' describing the relationship between baseline risk and risk under treatment is linear, i.e. 'relative risk' is constant. Absolute benefit is then proportional to a patient's baseline risk and the treatment is most effective among high-risk patients. Alternatively, the 'effect model' becomes curvilinear when 'odds ratio' is considered to be constant. However these two models are based on purely empirical considerations, and there is still no theoretical approach to support either the linear or the non-linear relation.

### Presentation of the hypothesis

From logistic and sigmoidal Emax (Hill) models, we derived a phenomenological model which includes the possibility of integrating both beneficial and harmful effects. Instead of a linear relation, our model suggests that the relationship is curvilinear i.e. the moderate-risk patients gain most from the treatment in opposition to those with low or high risk.

### Testing the hypothesis

Two approaches can be proposed to investigate in practice such a model. The retrospective one is to perform a meta-analysis of clinical trials with subgroups of patients including a great range of baseline risks. The prospective one is to perform a large clinical trial in which patients are recruited according to several prestratified diverse and high risk groups.

### Implications of the hypothesis

For the quantification of the treatment effect and considering such a model, the discrepancy between odds ratio and relative risk may be related not only to the level of risk under control conditions, but also to the characteristics of the dose-effect relation and the amount of dose administered. In the proposed approach, OR may be considered as constant in the whole range of *Rc*, and depending only on the intrinsic characteristics of the treatment. Therefore, OR should be preferred rather than RR to summarize information on treatment efficacy.

## Background

*Rc*) and on the y axis the risk of event under treatment (

*Rt*) (see Figure 1). The identity line corresponds with no treatment effect. For the dots below this line,

*Rt*is lower than

*Rc*and the treatment is beneficial. While for those falling above this line, the treatment seems unfavorable. A regression line may be estimated assuming the relative risk (RR) is constant, whereas absolute risk reduction (ARR) varies, possibly almost proportionately, with the baseline risk of patients. However, several problems accompany the linear relation between

*Rc*and

*Rt*: firstly, the linear model is purely empirical ; secondly, it does not take into account the range of variation of

*Rc*and

*Rt*, limited from 0 to 1 ; lastly, the validity of linear extrapolation for the patients in high risk (i.e.

*Rc*> 0.5) remains most often unknown.

A simulation approach [7] based on numerical models of drug action has explored the whole range of *Rc* and suggested a curvilinearity at least in some settings. However, there is still no theoretical approach to support such a non-linear relation. Consequently, we propose an alternative approach to revisit the relationship between *Rc* and *Rt*, based on a phenomenological model that includes the possibility of integrating both beneficial and harmful therapeutic effects.

## Presentation of the hypothesis

Where *R* corresponds to the probability of the outcome, *β*_{0} and *β*_{1} are the intercept and slope of linear function respectively, and E is the pharmacodynamic effect (or similarly a risk factor of the spontaneous outcome).

where E is the pharmacodynamic effect, E_{0} represents baseline value of E, E_{max} is the maximum theoretical effect, D is the dose, ED_{50} is the dose at which 50% of the maximum effect is achieved, and *γ* is the sigmoidicity parameter.

*Rt*and the treatment parameters are linked by :

*Rc*is given by

a model with *Rc* as independent variable, *Rt* as dependent variable, and *β*_{1}, E_{max}, ED_{50}, D, and *γ* as parameters.

Therefore, OR which represents the size of treatment efficacy, does not vary with *Rc*. OR is here a constant only determined by pharmacotherapeutic parameters. This remains also true in case of an active control therapy (D ≠ 0).

*Rc*and

*Rt*or ARR (left and right column respectively), assuming OR as a constant. In Figure 2B (left column), the quantitative treatment effect is shown as a function of the dose D. This effect is expressed either as an OR (thick line whatever the level of

*Rc*) or a RR (thin lines for increasing levels of

*Rc*). Compared with RR, the difference in the estimation of the treatment effect using OR is greater for high levels of

*Rc*, and also depends on the value of the administered dose D. The same results are shown in Figure 2B (right column) using ARR as the expression of the quantitative treatment effect. Additional simulations could be performed considering the other parameters of the dose-effect relation (i.e. E

_{max}, ED

_{50}and

*γ*).

_{1}) or harm (OR

_{2}) can be expressed with its own pharmacological parameters as:

where *i* = 1 for benefit and *i* = 2 for harm.

*P*=

*P*

_{1}+

*P*

_{2}-

*P*

_{1}·

*P*

_{2},

*Rt*and

*Rc*can be written by :

Then from models (7) and (8), for a treatment with one or two effects respectively, *Rt* and other measures of treatment effect, such as RR and ARR, can be easily translated and the relationships between baseline risk and treatment effect can be graphically simulated over the whole range of *Rc* (Figure 2C). The U-shape relation between *Rc* and *Rt* indicates that the patients with moderate risk obtain most from the treatment while those at highest risk benefit less or not at all. For a treatment involving two independent mechanisms, a harmful effect is observed in low-risk groups, and treatment effects under different doses show the potential difficulties of choosing an optimal dose for a patient, especially for those in low baseline risk.

## Testing the hypothesis

Two approaches (retrospective and prospective) can be proposed in order to investigate in practice such a non-linear 'effect model'. The retrospective one is to perform a meta-analysis of clinical trials with subgroups of patients with a great range of baseline risks. An alternative is to conduct a meta-analysis of individual patient data, a practical way of detecting differential treatment effects among various risk groups. [10] The prospective approach is to perform a large clinical trial ('megatrial') or several trials, in which patients are recruited according to several diverse risk groups that are prestratified depending on the information taken from previous small trials [11]. Severe patients, who are usually excluded from clinical trials, will be then preferentially included to detect the real treatment effect in high-risk conditions.

## Implications of the hypothesis

Using such a phenomenological approach based on the logistic and sigmoidal Emax (Hill) models, we observe a non-linear relation between *Rc* and *Rt*, that confirms previous results. [7, 12]

These results are however strongly dependent on the underlying models considered. Logistic regression, one of a class of models known as generalized linear models, is a type of predictive model that can be used with two types of target variables : a categorical one that has exactly two categories (i.e., a *binary* or *dichotomous* variable), or a continuous one that has values in the range 0 to 1 representing probability values or proportions; explanatory variables can be either categorical or quantitative. The log odds is then a linear function of the explanatory variable(s), leading to an OR constant. This model is now used extensively in the medical science, where a clinical outcome is in most cases binary (e.g. death/alive, event/no event). The dose-concentration-effect model (e.g. Hill model) is based on the law of mass-action and on receptor occupancy theory, supposing a reversible drug pharmacodynamic effect which is related to the clinical (binary) outcome. Depending on the biological effects of drugs and their mechanisms of action, such assumptions may not be always valid, particularly in case of irreversible mechanisms, tolerance/rebound phenomena, and synergistic/antagonist effects.

For most patients (e.g. those in low or moderate risk), one can assume a linear relation with an absolute benefit proportional to the baseline risk. However for high-risk patients, appropriate data are required to show up whether it remains proportional or decreases to zero. The values of OR and RR may consequently differ according not only to the level of risk under control conditions (*Rc*), but also to the characteristics of the dose-effect relation and the amount of the dose chosen. In the proposed approach, OR may be considered as constant in the whole range of *Rc*, and depending only on the intrinsic characteristics of the treatment. Therefore, OR should be preferred to summarize information on treatment efficacy.

## Declarations

### Acknowledgements

We are grateful to Michel Cucherat, and Michel Lièvre, for helpful discussions.

## Authors’ Affiliations

## References

- Collet JP, Boissel JP:
**Sick population-treated population: the need for a better definition.***The VALIDATA Group. Eur J Clin Pharmacol.*1991, 41: 267-71.View ArticlePubMed - Boissel JP:
**Individualizing aspirin therapy for prevention of cardiovascular events.***JAMA.*1998, 280: 1949-50.View ArticlePubMed - Boissel JP, Cucherat M, Gueyffier F:
**What target for guidelines: uniformed or tailored?***Am J Hypertens.*1998, 11: 890-3.View ArticlePubMed - Arends LR, Hoes AW, Lubsen J, Grobbee DE, Stijnen T:
**Baseline risk as predictor of treatment benefit: three clinical meta-re-analyses.***Stat Med.*2000, 19: 3497-518.View ArticlePubMed - Boissel JP, Collet JP, Lièvre M, Girard P:
**An effect model for the assessment of drug benefit: example of antiarrhythmic drugs in post-myocardial infarction patients.***J Cardiovasc Pharmacol.*1993, 22: 356-363.View ArticlePubMed - L'Abbe KA, Detsky AS, O'Rourke K:
**Meta-analysis in clinical research.***Ann Intern Med.*1987, 107: 224-33.View ArticlePubMed - Boissel JP, Cucherat M, Nony P, Chabaud S, Gueyffier F, Wright JM, Lièvre M, Leizorovicz A:
**New insights on the relation between untreated and treated outcomes for a given therapy : effect model is not necessarily linear.***J Clin Epidemiol.*2008, 61: 301-7.View ArticlePubMed - Bagley SC, White H, Golomb BA:
**Logistic regression in the medical literature: standards for use and reporting, with particular attention to one medical domain.***J Clin Epidemiol.*2001, 54: 979-85.View ArticlePubMed - Gabrielsson J, Weiner D:
**Pharmacokinetic and Pharmacodynamic Data Analysis.***Concepts and Applications.*2000, chapter 4: 177-189. Swedish Pharmaceutical Press, 3. - Stewart LA:
**Practical methodology of meta-analyses (overviews) using updated individual patient data.***Statist Med.*1995, 14: 2057-79. 10.1002/sim.4780141902.View Article - Ioannidis JPA, Lau J:
**The impact of high-risk patients on the results of clinical trials.***J Clin Epidemiol.*1997, 50: 1089-98.View ArticlePubMed - Deeks JJ:
**Issues in the selection of a summary statistic for meta-analysis of clinical trials with binary outcomes.***Statist Med.*2002, 21: 1575-1600. 10.1002/sim.1188.View Article

## Copyright

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.