Multifactorial assessment of the risk in periodontal diseases and dental implant management

A typical question of researchers and clinicians is : What is the relationship of one or more exposure variables to a disease outcome ? In the advent of the third millennium, the answer to this important question has completely changed our understanding of the pathogenesis of multifactorial diseases and, consequently, the clinical management of patients who accumulate these variables. These patients are now named " risk patients " whereas the variables are so called " risk factors ". ^{Beck (1994)} has defined a risk factor as part of the causal chain for a particular disease or a feature (i.e., personal behavior, environmental exposure or genetic characteristics) which can expose the patient to this disease. This causal relationship has direct implications in the disease probability which can be increased or decreased according to the presence or the absence of the variables.

So far, it was not possible to accurately quantify this probability and thus identify risk patients. Statistical models in research allowing adjustment for co-risk factors and confounders provide new information able to shift the clinical management of infectious diseases. Thanks to the statistics and risk assessment studies, it is possible for the clinician to answer the question : how many times is the risk of a particular patient contracting or developing the disease increased, mindful that he is exposed to one or multiple risk factors ? Furthermore, if this disease can be considered as a risk factor, it itself becomes a variable for another disease and its influence on the occurrence of the latter can be also quantified.

Thus, the statistical approach to risk is the keystone in our knowledge of the failure or sucess probabilities of our treatments. The new concept is that risk factors have to be evaluated in our patient's prognosis and treatment.

Basic statistical approach of risk

From a medical point of view, risk is the probability of an individual developing a given disease or a pathological change over a specific period (^{Kleinbaum et al., 1982}). Three classes of variables may be selected to assess their relationship to a multifactorial disease (^{table I}).

Different types of studies for assessing risk are at the researcher's disposal. Case-reports suggest risk association. Case-control studies identify risk indicators. Higher levels of evidence in the identification of risk indicators are given by cross-sectional studies. Longitudinal epidemiologic studies identify true risk factors. Intervention studies, i.e., randomized clinical trials, enable to test the effects of modifying risk factors on the disease. Stricly speaking, only a factor confirmed by a randomized clinical trial should be labelled as a « risk factor ».

Univariate analysis

This is the analysis of the association of two variables. If we consider a dichotomous disease outcome which might be, for example, periodontitis or peri-implantitis, with subjects being classified as either D0 (no disease) or D1 (presence of the disease), it can be of interest to evaluate the extent to which smoking (S1) or not smoking (SO) is associated with periodontal or peri-implant status. This may be summarized in a 2 x 2 table (^{table II}).

For example, a is the number of smokers who have the disease in the sample. The sample size is n = a + b + c + d.

R1 is the risk of having the disease (D1) when you are a smoker (S1) and R0 the risk of having the disease when you do not smoke (S0). One way of presenting the effect of smoking on the disease would be as the absolute difference between these two risks, known as the absolute risk (R).

R = R1 - R0 = a/(a+b) - c/(c+d)

Another way of expressing the impact of smoking would be as a relative risk or risk ratio (RR).

RR = R1 / R0 = (a/(a+b)) / (c/(c+d))

However, this relative risk can only be used if the data are derived from a follow-up study, whereas the odds ratio (OR) is the only directly estimated measure of association, regardless of whether the study design is follow-up, case-control, or cross-sectional.

OR = (R1/(1-R1)) / (R0/(1-R0)) = ad / bc

There is no unit to express RR or OR. For example, if OR = 2.5 this means that there is 2.5 times more risk of smokers contracting the disease than non smokers. To summarize the RR or OR value :

> 1 : variable must be considered as a risk factor

= 1 : variable must not be considered as a risk factor

< 1 : variable has a protective effect against the disease

Multivariate analysis

This allows the analysis of multiple risk factors and their adjustment for various variables. So far, we have not paid attention to additional variables, such as age, race, plaque levels or sex, which were not of primary interest in our hypothetical study. However, it is important to adjust the smoking-disease association to these variables. The exposure variable (smoking) together with the control variables (age, race, plaque levels, sex) represent a collection of independent variables that we wish to use to predict the dependent variable (disease). We have a flexible choice for these variables which can be exposure, control or even combinations of both.

Thus, we are considering a multivariable problem. This complex inter-relationship between many variables requires the use of a mathematical model such as logistic regression. Based on the data from an epidemiologic study, this model identifies the risk factors from this study and predicts the risk of the disease for a specific patient.

Logistic regression is the most popular multivariate procedure used to analyze and select epidemiologic data when the disease is a dichotomous variable (i.e. having or not having the disease). It can be used for any dichotomous (sex, for example) or continuous (age, for example) control variables. The model's formula is :

P(X) = 1 / (1 + exp ( - a + S b_i x X_i) )

X_i is the collection of variables measured at different times, α and β _i represent unknown parameters which are given through data of the study by a method of estimation. They are fixed parameters whereas it is possible to modify the variables : 0 or 1 for dichotomous variables and 1 to 90 for example for a continuous variable such as age. Thus, the P(X) value will be different according to changes in the control variable.

Suppose we want to use this model to obtain the predicted risk for a specific patient, namely to estimate the probability of having the disease. He is 20, male and a smoker. Calculation of P(X) will give, for example, a 20 % risk. If now we replace the 1 value of smoking by the 0 value of non smoking the P(X) will change and will give, for example, a 5 % risk. If we divide the predicted risk of the smoker by that of the non smoker, we get the risk ratio (RR).

RR = 0,20 / 0,05 = 4

This means that this patient has 4 times more risk of having the disease than a non smoker with the same age and the same sex. This conclusion cannot be drawn without the confidence interval (CI).

If the confidence interval of the RR or the OR contains 1 (for example, CI = [0.91 - 1.08]),the comparison smoker vs non smoker is not statistically significant (CI_{95 %}).

Logistic regression is robust and can be applied to any study designs, including not only follow-up investigations but also case control and cross sectional studies. However, odds ratios can only be obtained for case-control and cross-sectional studies. In these studies, risk estimates cannot be estimated. If the aim of the study is to evaluate the exposure-disease association in terms of an odds ratio, this limitation is not severe. Nevertheless, the odds ratio provides a good approximation to the risk ratio. If the disease is rare, and this is the case for peri-implantitis and certain forms of periodontitis, it can be assumed that the OR estimate from the logistic regression model approximates the RR.

It has been stressed that a distinction has to be made between risk predictors and true risk factors (^{Papapanou and Lindhe, 1997}). Prognosis factors are related to disease progression whereas true risk factors are related to the onset of the disease. This academic discussion is important because, as described above, the multivariate analysis may include a collection of variables which can interact with each other. The inclusion of a variable which is a consequence of the disease can be a powerful predictor and, consequently, may mask the emergence of a true risk factor.

Conclusion

A new field of investigation in periodontal research has been developed as we now know that periodontitis are bacterial infections and multifactorial diseases. In many infectious diseases, the sole presence of the causative microbial agent is not sufficient to trigger a pathological process. Periodontal diseases are in the scope of this new concept of bacterial disease development. Some studies estimate that bacteria account only for 20 % of the variance in susceptibility for disease expression (^{Grossi et al., 1994}) whereas hereditary factors should account for around 50 % (^{Michalowicz, 1993}).

Due to the small number of longitudinal studies involving multivariate analysis of the data and the subsequent randomized clinical trials, identification of risk factors is a newborn challenge which will take time to complete. Smoking and diabetes seem to be true risk factors. Stress and osteoporosis are potential risk factors. There is not enough evidence to consider other variables as true risk factors. Regarding implant therapy, studies are lacking.

Thus, the clinical management of periodontal disease by physical disruption of the microbial biofilm remains the crucial component of periodontal therapy (^{Page et al., 1997}). However, we must keep in mind that, as we know little about true risk factors, all variables able to increase the risk of a pathogenic process must be taken into account for prognosis evaluation with or without treatment.

The author wishes to thank Roger Caillon and Raneesa Kassim-Premdjee for their advice in writing this article.

BIBLIOGRAPHY

• BECK JD. Methods of assessing risk for periodontitis and developing multifactorial models. J Periodontol 1994;65:468-478.
• GROSSI SG, ZAMBON JJ, HO AW, KOCH G, DUNFORD RG, MACHTEI EE, NORDERYD OM, GENCO RJ. Assessment of risk for periodontal disease. 1. Risk indicators for attachment loss. J Periodontol 1994;65:260-267.
• KLEINBAUM D, KUPPER L, MORGENSTERN H. Epidemiologic research : principles and quantitative methods. Belmont, CA : Liftime Learning Publications, 1982 : 3.
• MICHALOWICZ BS. Genetic and inheritance considerations in periodontal disease. Curr Opin Periodontol 1993;1:11-176.
• PAGE RC, OFFENBACHER S, SCHROEDER HE, SEYMOUR GJ, KORNMAN KS. Advances in the pathogenesis of periodontitis : summary of developments, clinical implications and future directions. Periodontology 2000 1997;14:216-248.
• PAPAPANOU PN, LINDHE J. Epidemiology of periodontal disease in Clinical Periodontology and Implant Dentistry. Lindhe J, Karring T, Lang NP, eds. Copenhagen : Munksgaard, 1997:69-101.