HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Figures Only Full Text (PDF) Services Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via ISI Web of Science (27) Citing Articles via Google Scholar Google Scholar Articles by Spiekerman, C.F. Articles by DeRouen, T.A. Search for Related Content PubMed PubMed Citation Articles by Spiekerman, C.F. Articles by DeRouen, T.A.

Bias Induced by Self-reported Smoking on Periodontitis-Systemic Disease Associations

¹ Department of Dental Public Health Sciences, Box 357475, University of Washington School of Dentistry, Seattle, WA 98195;
² Department of Epidemiology, University of Washington School of Public Health;
³ Department of Biostatistics, University of Washington School of Public Health;

^* corresponding author, cspieker{at}u.washington.edu

	ABSTRACT

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

 
Non-causal associations between periodontitis and systemic diseases may be spuriously induced by smoking because of its strong relationship to both. The goal of this study was to evaluate whether adjustment for self-reported smoking removes tobacco-related confounding and eliminated such spurious confounding. Using NHANES III data, we evaluated associations between attachment loss and serum cotinine after adjustment by self-reported number of cigarettes smoked. Cotinine, a metabolite of nicotine, should not be related to attachment loss, if self-reported smoking captures the effect of tobacco on attachment levels. Adjustment for self-reported cigarette smoking did not completely remove the correlation between attachment loss and serum-cotinine level (r = 0.075, n= 1507, p = 0.003). Simulation studies indicated similar results for time-to-event data. These findings demonstrate the difficulty in distinguishing the effects of periodontitis from those of smoking with respect to a smoking-related outcome. Future studies should report results of analyses on separate subcohorts of never-smokers and smokers.

KEY WORDS: periodontitis • smoking • systemic disease • confounding • NHANES III

	INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

 
A current controversial topic in the medical-dental literature concerns possible causal associations between periodontal disease and a wide array of tobacco-related systemic diseases. Reports have indicated periodontal disease to be associated with heart disease (Mattila et al., 1989; DeStefano et al., 1993; Beck et al., 1996), cerebrovascular disease (Wu et al., 2000), adverse pregnancy outcomes (Offenbacher et al., 1996), diabetes (Grossi and Genco, 1998), and lung cancer (Hujoel et al., 2003). Because periodontal disease is itself heavily influenced by cigarette usage (Haber et al., 1993; Tomar and Asma, 2000), controlling for the confounding effects of smoking is prerequisite to credible claims of causality in these associations. Unfortunately, statistical adjustment for tobacco exposure has proven to be a complicated issue. Because of the imprecision of smoking data gleaned from self-report, it is difficult to remove fully the confounding effects of true tobacco exposure. In a situation such as this, where cigarette smoking is such an important risk factor for both associated diseases, there is potential for the residual confounding to distort estimates of association significantly.

Error in the measurement of confounder variables hinders the efficacy of standard statistical-modeling techniques in removal of the confounding effects. The resulting residual confounding may produce bias in the estimates of association between an exposure and outcome. In the setting of linear regression, one can show that the magnitude of the bias will be positively related to the amount of error involved in measurement of the confounder, the association between the confounder and the exposure, and the association between the confounder and the outcome, and inversely related to the variations of the confounder and of the outcome. In estimates of associations between periodontitis and smoking-related systemic diseases, the first three of these components are large. Measurements of the confounder, tobacco exposure, are usually derived from participant self-reports, which have been reported to be suspect in terms of the numbers of cigarettes smoked per day (Klesges et al., 1995; Gonzalez et al., 1996). Tobacco exposure is a very important risk factor for periodontitis and is also strongly implicated in, if not the greatest risk factor for, many systemic diseases currently linked with periodontitis. Rarely has there been a situation where a confounder has such a strong relation to both exposure and outcome. Combine this with the difficulties inherent in ascertaining accurate smoking information and one has a situation in which estimates of association may be heavily influenced by residual confounding.

To get a clear idea of the effects of typical inaccuracies of self-reported smoking data on estimates of association in periodontal-systemic disease investigations, we examined relationships between measures of periodontal disease and levels of serum cotinine. Cotinine, the major metabolite of nicotine, is an objective measure of current tobacco exposure (Hill et al., 1983). Serum cotinine levels are clearly a consequence of tobacco exposure. At the same time, there is no reason to believe them to be in any way affected by the existence or the severity of periodontal disease. In this article, we use data from the Third National Health and Nutrition Examination Survey (NHANES III) to demonstrate that adjustment for tobacco exposure using the imperfect measure of self-reported smoking is not sufficient to remove completely the confounding effects of smoking. We present the results of standard statistical analyses in which false-positive associations are indicated between measures of periodontal disease and outcomes determined completely by serum cotinine, even after adjustment for tobacco exposure based on self-reported smoking data.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

 
The data used were from the NHANES III, the Third National Health and Nutrition Examination Survey. NHANES III, conducted in two stages from 1988–1991 and 1991–1994, is the latest of a series of large surveys conducted by the United States Department of Health and Human Services with the purpose of evaluating overall health and nutrition characteristics of non-institutionalized individuals in the United States (Centers for Disease Control and Prevention, 1994; NHANES, 1996). The survey included a comprehensive clinical dental examination component. NHANES III was approved by the Institutional Review Board of the National Center for Health Statistics.

In the NHANES III dental exam, loss of attachment was measured at the buccal and mesial-buccal aspects of all teeth from randomly assigned half-mouths (one upper and one lower quadrant), excluding third molars. Serum cotinine was assayed from samples collected via venipuncture.

We used data from the 1507 participants who had data available on attachment loss and cotinine level, who self-reported current smoking of at least one cigarette per day, and who did not report current use of cigars, pipes, chewing tobacco, or snuff. For each participant, we computed the mean attachment loss by taking the average of all available loss-of-attachment measurements.

The coefficient of partial correlation and a partial regression plot (Neter et al., 1990) were computed to present the association remaining between attachment loss and cotinine after adjustments for tobacco exposure, age, race, and gender. We adjusted for tobacco exposure using the logarithm of the self-reported number of cigarettes smoked per day, since the association between serum cotinine levels and cigarettes smoked per day appeared to have a logarithmic rather than a linear shape. We also adjusted for age (age and age squared), race (white, other), and gender in an attempt to reduce confounding unrelated to smoking.

To assess how the residual confounding of tobacco exposure may influence estimates of association in standard time-to-event (survival) analyses, we performed computer-simulation experiments. Sampling, with replacement, from the 1507 participants, we generated datasets with joint distributions of cotinine, self-reported smoking, attachment loss, age, gender, and race, equivalent to those in our NHANES III cohort. We then randomly generated variables to simulate times to fictional morbid events. Times-to-events were specified according to the formula t = X•exp(-ß•cotinine/287), where t denotes the time-to-event, X is an exponential random variable with mean 1, and exp(ß) specifies the relative risk of event related to a 287 ng/mL difference in serum cotinine. The figure 287 ng/mL is the average serum cotinine level for those 373 participants who reported smoking 20 cigarettes (one pack) per day. Note that the mechanism generating the times-to-events was based entirely on serum cotinine and did not utilize attachment loss. Each simulation study had a fixed censoring time that could be thought of as the time of a study after which no more data are collected. For each computer-generated observation, the length of follow-up was recorded to be the minimum of the time-to-event and the censoring time. If the time-to-event was shorter than the censoring time, then an ‘observed’ event was indicated. For each simulation study, the fixed censoring time was set to the time that would produce specific cumulative incidences of observed morbid events. Using these methods, we generated datasets for simulation studies using sample sizes n = 5000 and n = 10000 and morbid event cumulative incidences 5% and 25%. The simulation datasets were computed with the use of relative risks of morbid events associated with cotinine of exp(ß) = 1.9 and exp(ß) = 3.8, risk estimates of CHD reported to be associated with the smoking of 20 cigarettes/day among males and females, respectively (Woodward et al., 1999). For each simulation study, 5000 datasets were generated.

For each dataset, we used Cox proportional-hazards models to estimate associations between the morbid event and mean attachment loss. We adjusted for the confounding effect of tobacco exposure via two methods. First, we adjusted using level of serum cotinine. This method was considered ‘perfect’ adjustment, since serum cotinine values were the exact values used to produce the computer-generated morbid events. The second method of adjustment used the logarithm of self-reported number of cigarettes smoked as the tobacco-adjustment variable. Each model was also adjusted for age, race, and gender.

For each simulation study, median relative-risk estimates were tabled as well as the proportions of datasets for which the Cox-regression models indicated statistically significant (p < 0.05) associations between the periodontal disease measures and hazard of event. We used the S-plus 2000 computer language to compute the simulation studies.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

 
The average (± standard deviation) mean attachment loss of the 1507 participants was 1.53 ± 1.42 mm. The mean (± standard deviation) reported numbers of cigarettes smoked per day and levels of serum cotinine were 15.6 ± 11.2 cigarettes and 225 ± 150 ng/mL, respectively. The unadjusted Pearson correlations of cotinine with log cigarettes smoked and mean attachment loss were r = 0.500 and r = 0.173, respectively.

With the original data, a linear regression model adjusting for log self-reported cigarettes smoked per day, age, gender, and race indicated mean attachment loss to be statistically significantly associated with serum cotinine (partial correlation coefficient = 0.075, p = 0.003). The partial regression plot presented in the Fig. depicts the linear association remaining between attachment loss and cotinine after adjustments.

View larger version (25K): [in this window] [in a new window]   Figure. Partial regression plot of 1507 current smokers in the NHANES III study, depicting the linear association remaining between mean attachment loss and serum cotinine after adjustment for log of self-reported cigarettes smoked per day, age, age squared, race (white, other), and gender.

We also ran simulations using a dichotomous indicator of periodontitis (mean attachment loss > 5 mm) as the definition of periodontal disease. Imperfect adjustment for tobacco exposure resulted in median relative risks ranging from 1.03 to 1.10. The percentages of datasets for which significant associations were indicated ranged from 5.1% to 12.2% (data not shown).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

 
Standard regression methods rely on the assumption that the independent variables are measured without appreciable error. The use of an imperfect measure like self-reported smoking to adjust for tobacco exposure will preclude full adjustment for the effects of smoking. The ‘leftover’ confounding due to the inadequate adjustment may affect estimates of association for any of the variables in the model. The effect is analogous to that of leaving out an important confounder variable. If both the outcome and the exposure of interest are positively correlated with the confounder (as is the case in an investigation of the relationship between periodontitis and a tobacco-related systemic disease), any residual confounding will falsely elevate the estimates of association between exposure and outcome. This report demonstrates that the smoking-periodontitis correlations and the self-reported smoking data imprecisions, found in a real-life well-known dataset, are of sufficient magnitude to produce false-positive associations between measures of periodontitis and a smoking-related condition.

For a poorly measured confounder variable to contribute to residual confounding, it is not necessary that the misclassification be all in one direction (e.g., consistent under-reporting) or depend upon the exposure of interest. Any misclassification that results in study participants being classified as similar when, in fact, they significantly differ in terms of a true confounder will result in uncontrolled confounding. For example, adjusting for smoking using a dichotomous smoker/non-smoker variable will classify all smokers into the same group. Because the risks of periodontal disease and, say, stroke are both increased by amount of smoking, an analysis that adjusts for smoking using only smoker/non-smoker will leave substantial confounding unadjusted for among the group of smokers.

A better and probably the best measure of current tobacco exposure available in most studies is the self-reported number of cigarettes smoked per day. Unfortunately, the utility of this finer measure will be limited by the abilities of participants to gauge their true intake accurately. The misclassification resulting from unintentional or intentional misreporting, differences in amounts of inhalation, differences in toxicity of different types of cigarettes, etc., will still seriously hinder any efforts to adjust completely and fully for the effects of smoking. Because smoking is so strongly correlated to severity of periodontal disease and the risk of many systemic diseases, a possibly small level of misclassification can still result in significant distortions of estimates of association.

Possible good news is that the magnitudes of the median biases appear to be rather small in the simulation results. However, it is difficult to extrapolate exact magnitudes from these fairly simple simulations based on cotinine to studies involving actual systemic diseases generated via varied and complicated mechanisms. It is safe to conclude that the error found in self-reported smoking data is producing some bias which is elevating risk estimates associating periodontitis with smoking-related diseases. Because of the small magnitudes of most currently posited associations, identifications of even small biases are important to consider.

A method that shows promise for evaluating the magnitude of the bias induced by smoking is stratification of the cohort by smokers (former and current) and never-smokers, and comparison of the estimates of systemic-disease-periodontitis associations between the two cohorts. Because adjustment for smoking levels will be a non-issue among never-smokers, one would expect biases caused by insufficient adjustment for smoking levels to be present only among current and former smokers. Several reports (Hujoel et al., 2001, 2002; Hyman et al., 2002) have found a greater association between periodontitis and systemic disease among smokers as compared with never-smokers. Hyman et al. (2002) posit that these differences indicate smoking to be a necessary co-factor for periodontitis to induce systemic damage. The results of the current study indicate that residual confounding due to smoking is also a credible explanation. Whichever the reason for the differences, evaluating associations in these two cohorts separately appears to be a valuable exercise.

It is important to note that this report considers only the confounding linked specifically to current smoking levels. Additional residual confounding may result from factors such as the difficult-to-quantify socio-economic status, or smoking "packyears", which often incorporates self-reported smoking rate into its calculation. Residual confounding from these other factors would likely add to the false elevation of periodontitis-systemic disease association estimates.

Periodontitis is profoundly influenced by cigarette smoking. The systemic diseases currently linked to periodontitis are similarly influenced by exposure to tobacco smoke. The analyses presented here have demonstrated that statistical adjustment using self-reported smoking data may not be sufficient to remove completely the confounding effects of current tobacco exposure. It is plausible that bias resulting from such residual confounding is at least partially responsible for some of the many low-level associations between periodontitis and various smoking-related systemic diseases reported in the current literature.

	ACKNOWLEDGMENTS

 
The NHANES III data were provided by the National Center for Health Statistics. This study is supported by NIH/NIDCR P30 DE09743 and R03-DE13861.

Received January 2, 2002; Last revision January 8, 2003; Accepted January 29, 2003

	REFERENCES

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

 
Beck J, Garcia R, Heiss G, Vokonas PS, Offenbacher S (1996). Periodontal disease and cardiovascular disease. J Periodontol 67(Suppl 10):1123–1137.[ISI][Medline]

Centers for Disease Control and Prevention (1994). Plan and operation of the Third National Health and Nutrition Examination Survey, 1988–94. Vital Health Statistics 1(32). Series 1, No. 32. Bethesda, MD: National Center for Health Statistics.

DeStefano F, Anda RF, Kahn HS, Williamson DF, Russell CM (1993). Dental disease and risk of coronary heart disease and mortality. BMJ 306(6879):688–691.

Gonzalez YM, De Nardin A, Grossi SG, Machtei EE, Genco RJ, De Nardin E (1996). Serum cotinine levels, smoking, and periodontal attachment loss. J Dent Res 75:796–802.[Abstract/Free Full Text]

Grossi SG, Genco RJ (1998). Periodontal disease and diabetes mellitus: a two-way relationship. Ann Periodontol 3:51–61.[Medline]

Haber J, Wattles J, Crowley M, Mandell R, Joshipura K, Kent RL (1993). Evidence for cigarette smoking as a major risk factor for periodontitis. J Periodontol 64:16–23.[ISI][Medline]

Hill P, Haley NJ, Wynder EL (1983). Cigarette smoking: carboxyhemoglobin, plasma nicotine, cotinine and thiocyanate vs self-reported smoking data and cardiovascular disease. J Chronic Dis 36:439–449.[ISI][Medline]

Hujoel PP, Drangsholt MT, Spiekerman C, DeRouen TA (2001). Periodontal disease and risk of coronary heart disease. J Am Med Assoc 285:40–41.[Free Full Text]

Hujoel PP, Drangsholt M, Spiekerman C, DeRouen TA (2002). Periodontitis-systemic disease associations in the presence of smoking—causal or coincidental? Periodontol 2000 30:51–60.

Hujoel PP, Drangsholt MT, Spiekerman C, Weiss NS (2003). An exploration of the periodontitis-cancer association. Ann Epidemiol (in press).

Hyman JJ, Winn DM, Reid BC (2002). The role of cigarette smoking in the association between periodontal disease and coronary heart disease. J Periodontol 73:988–994.[ISI][Medline]

Klesges RC, Debon M, Ray JW (1995). Are self-reports of smoking rate biased? Evidence from the Second National Health and Nutrition Examination Survey. J Clin Epidemiol 48:1225–1233.[ISI][Medline]

Mattila KJ, Nieminen MS, Valtonen VV, Rasi VP, Kesaniemi YA, Syrjala SL, et al. (1989). Association between dental health and acute myocardial infarction. BMJ 298(6676):779–781.

Neter J, Wasserman W, Kutner MH (1990). Applied linear statistical models. 3rd ed. Homewood, IL: Irwin.

Offenbacher S, Katz V, Fertik G, Collins J, Boyd D, Maynor G, et al. (1996). Periodontal infection as a possible risk factor for preterm low birth weight. J Periodontol 67:1103–1113.[ISI][Medline]

The Third National Health and Nutrition Examination Survey (NHANES III 1988–94) (1996). Reference Manuals and Reports [book on CD-ROM]. Bethesda, MD: National Center for Health Statistics. http://www.cdc.gov/nchs/nhanes.htm.

Tomar SL, Asma S (2000). Smoking-attributable periodontitis in the United States: findings from NHANES III, National Health and Nutrition Examination Survey. J Periodontol 71:743–751.[ISI][Medline]

Woodward M, Moohan M, Tunstall-Pedoe H (1999). Self-reported smoking, cigarette yields and inhalation biochemistry related to the incidence of coronary heart disease: results from the Scottish Heart Health Study. J Epidemiol Biostat 4:285–295.[Medline]

Wu T, Trevisan M, Genco RJ, Dorn JP, Falkner KL, Sempos CT (2000). Periodontal disease and risk of cerebrovascular disease: the first national health and nutrition examination survey and its follow-up study. Arch Intern Med 160:2749–2755.[Abstract/Free Full Text]

This article has been cited by other articles:

				Y.-K. Tu, B. Galobardes, G. D. Smith, P. McCarron, M. Jeffreys, and M. S Gilthorpe Associations between tooth loss and mortality patterns in the Glasgow Alumni Cohort Heart, September 1, 2007; 93(9): 1098 - 1103. [Abstract] [Full Text] [PDF]

				D.J. Caplan, J.B. Chasen, E.A. Krall, J. Cai, S. Kang, R.I. Garcia, S. Offenbacher, and J.D. Beck Lesions of endodontic origin and risk of coronary heart disease. J. Dent. Res., November 1, 2006; 85(11): 996 - 1000. [Abstract] [Full Text] [PDF]

				M. L. Barnett and J. J. Hyman Challenges in interpreting study results: The conflict between appearance and reality J Am Dent Assoc, October 1, 2006; 137(suppl_2): 32S - 36S. [Abstract] [Full Text] [PDF]

				K. D Barnfather, G. F Cope, and I. L Chapple Effect of incorporating a 10 minute point of care test for salivary nicotine metabolites into a general practice based smoking cessation programme: randomised controlled trial BMJ, October 29, 2005; 331(7523): 999. [Abstract] [Full Text] [PDF]

				M.A. Fisher, G.W. Taylor, and K.R. Tilashalski Smokeless Tobacco and Severe Active Periodontal Disease, NHANES III J. Dent. Res., August 1, 2005; 84(8): 705 - 710. [Abstract] [Full Text] [PDF]

				Y.-K. Tu and M. S Gilthorpe Commentary: Is tooth loss good or bad for general health? Int. J. Epidemiol., April 1, 2005; 34(2): 475 - 476. [Full Text] [PDF]

				Th. Dietrich and K. Hoffmann A Comprehensive Index for the Modeling of Smoking History in Periodontal Research J. Dent. Res., November 1, 2004; 83(11): 859 - 863. [Abstract] [Full Text] [PDF]

				A. Taguchi, M. Sanada, Y. Suei, M. Ohtsuka, K. Lee, K. Tanimoto, M. Tsuda, K. Ohama, M. Yoshizumi, and Y. Higashi Tooth Loss Is Associated With an Increased Risk of Hypertension in Postmenopausal Women Hypertension, June 1, 2004; 43(6): 1297 - 1300. [Abstract] [Full Text] [PDF]

				M. T. Lydon-Rochelle, P. Krakowiak, P. P. Hujoel, and R. M. Peters Dental Care Use and Self-Reported Dental Problems in Relation to Pregnancy Am J Public Health, May 1, 2004; 94(5): 765 - 771. [Abstract] [Full Text] [PDF]

This Article Abstract Figures Only Full Text (PDF) Services Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via ISI Web of Science (27) Citing Articles via Google Scholar Google Scholar Articles by Spiekerman, C.F. Articles by DeRouen, T.A. Search for Related Content PubMed PubMed Citation Articles by Spiekerman, C.F. Articles by DeRouen, T.A.