Poor dental health and risk of pancreatic cancer: a nationwide registry-based cohort study in Sweden, 2009–2016

Background Previous studies have reported inconsistent results regarding the association between poor dental health and pancreatic cancer risk. This study aimed to assess this association using a well-functioning nationwide dental health registry in Sweden. Methods Information of exposures (dental caries, root canal infection, mild inflammation, and periodontitis; the number of teeth) was ascertained from the Swedish Dental Health Register, and occurrence of pancreatic cancer was identified from both cancer and cause of death registries. Hazard ratios (HRs) were estimated using Cox models. Results During a median of 7.2 years of follow-up, 10,081 pancreatic cancers were identified among 5,889,441 individuals. Compared with the healthy status, a higher risk of pancreatic cancer was observed in individuals with root canal infection, mild inflammation, and periodontitis in the <50 age group (P for trend <0.001). In the 50–70 age group, only the subgroup with periodontitis had an excess risk (multivariable-adjusted HR = 1.20, 95% confidence interval [CI] 1.11–1.29). No positive association with statistical significance was observed in the 70+ age group. Individuals with fewer teeth tended to have a higher risk in all age groups. Conclusions Our results confirmed the association between poor dental health and pancreatic cancer risk, which warrants further studies on underlying mechanisms.


The Swedish Dental Health Register (DHR)
The DHR has been described elsewhere [1]. Briefly, the DHR was initiated in July 2008, with good coverage from 2009, containing subjects' dental care information from both public and private dental care organizations in Sweden. This registry has high coverage of the Swedish adult population, with about 75% of individuals >19 years visiting dental clinics on a regular basis [2]. The DHR records clinical and administrative data, including number of teeth, diagnoses, dental procedures, and visit dates. However, only diagnoses that are subject to an intervention or treatment during the dental visit are recorded [1]. In general, an oral health examination is performed at a subject's first visit; if needed, further treatment will be performed in the following months. Thus, we used information of diagnoses and dental procedures from the first dental visit and all following visits within three months to establish a baseline diagnosis.

Exposures
For dental health status, when multiple diagnoses were present within the same date, the most severe according to the order of increasing severity was selected. Consequently, the most severe diagnosis across the first visit up until three months was ascertained as the baseline dental health status, and the date of first dental visit was set as date of entry into the cohort.
In addition, the information of number of teeth was matched to dental care data by the visiting dates. If a same-date record could not be found, then the information from the closest record up to six months before or five years after the first dental visit (entry date) was used. Additionally, records reported zero or 32 remaining and intact teeth were excluded, since we considered this type of record as unacceptably low quality based on the results in a validation study of the DHR [1].

Interactions for preliminary models
We estimated the interactions between age at baseline and poor dental health, and between age at baseline and number of teeth. To consider an interaction term, we multiplied the two terms together (continuous age*categorized dental health status / number of teeth), and compared the main effect model (not including the interaction term) with the interaction model (including the interaction term). A likelihood ratio test was used to assess the null hypothesis of no interaction.

Supplementary Discussion
Selection bias ─ Selection bias due to loss to follow-up is caused by the differences in completeness of follow-up between exposed and non-exposed groups, which is a common concern in cohort studies. In this study, we linked multiple nationwide registers with almost complete information from birth to death (i.e., migration, death, and outcome of interest), to minimize chances of loss to follow-up for each enrolled subject.
Reverse causality ─ Due to a lack of efficient diagnostic tools for pancreatic cancer, we were not able to identify early-stage pancreatic cancer or precursor lesions as the outcome of interest in the cohort. It implies that the observed exposure-outcome association is biased, due to the fact that the outcome may already exist before exposure occurs. Therefore, we added a lag time (one year) after the start of follow-up to minimize the risk of such bias. However, hazard ratios (HRs) were only marginally changed, which indicates a minimal impact from reverse causality.
Misclassification ─ It is possible that incident pancreatic cancer is underreported due to current diagnostic techniques and patient register system. In this study, pancreatic cancer cases were ascertained from both the cancer (incident cases) and cause of death registers (death cases) to estimate relative risks. HRs changed marginally when comparing the estimates based on incident pancreatic cancers only, with those combining incident and death cases, which indicates that underreporting of incident cancers does not strongly bias the results.
Surveillance bias ─ It may occur when some participants are followed up more closely and receive more surveillance, screening or tests than others, leading to an outcome diagnosed more frequently in these closely investigated individuals [3]. The general health care addresses broad health conditions and focuses on some specific areas of health if necessary, while dental care is limited to narrower field of health care: oral diseases and concerns. In this study, the relative risk estimates was only marginally changed when excluding the first year of follow-up, which indicates a minimal impact from surveillance bias. Therefore, this bias is likely to be limited.