Study design, setting and participants
This was a retrospective, unmatched case-control, hospital-based study that involved post-partum (PP) mothers in the maternity block at Muhimbili National Hospital (MNH), Dar-es-Salaam, Tanzania. The MNH is the largest hospital in Tanzania and handles referral patients from all district/municipal hospitals in Dar-es-Salaam region, nearby regional hospitals and other referral hospitals in the country. The PP mothers were those admitted at MNH for various problems of their newborn babies or themselves. The reasons for admission as inquired from mothers included prematurity, birth asphyxia, pneumonia, septicemia, jaundice, malaria, chorioamnionitis, gastro-intenstinal and umbilical cord problems, congenital malformations, anemia in pregnancy, uterus-cervical incompetence, and hypertension. The admissions of the newborns and their mothers were made between day 1 and 31 after delivery (mean 4.2 ± 6.3) with the median and mode being one. The time period that had elapsed between parturition and clinical intra-oral examination of the mother ranged from 1 to 40 days with a mean, median and mode of 8.1 ± 7.4, 5 and 2 days, respectively. The study participants that were examined between 1–7 days after delivery were 62.5%, 8–14 days (22.3%), 15–21 days (7.8%), 22–28 days (4.8%) and 29–40 days (2.7%)), however, the differences between cases and controls in the specified different times of examination were not statistically significant. The PP mothers were categorized into a "case" and "control" as follows: A "case" was defined as a PP mother who had at preterm (gestational age < 37 weeks) live low-birth-weight (PTLBW) singleton delivery weighing < 2,500 gms . Gestational age was based on last menstrual period instead of ultrasound uterine examination, which most mothers cannot afford. A "control" was defined as a PP mother who had at term (gestational age of ≥ 37 weeks) live normal-birth-weight singleton delivery (TNBW) weighing ≥ 2,500 gms. During the six months period of study, a total of 373 consecutive PP mothers (singleton normal deliveries) were recruited for this study (150 cases (PTLBW) and 223 controls (TNBW), a "case:control" ratio of about 1:1.5. Sampling of the participants was not done as all who consented and met the inclusion criteria, were incorporated in the study.
Inclusion and exclusion criteria
Eligibility to take part in the study was based on the inclusion criteria that qualified any PP mother who delivered a baby or being admitted with a PTLBW or TNBW infant at MNH, provided that the mother had no known underlying systemic disease(s). Screening for HIV was not performed. Exclusion criteria disqualified any PP mother that on medical records had infections or conditions other than periodontal diseases such as genitourinary tract infection, concurrent antibiotic therapy, diabetes mellitus, heart disease, glomerulonephritis, hyperthyroidism, HIV/AIDS, and those with a history of these diseases. This was deemed as necessary because all these conditions were considered to be confounding factors [5, 7, 10]. Mothers that had multiple pregnancies (twins, etc.) and/or delivered by caesarian-section were also excluded from the study. Eventually, a total of seventy-seven mothers (from both cases and control groups) were excluded from the present report due to the various reasons specified above including those that were "small for gestational age" (SGA), at term low birth weight (TLBW) and preterm normal birth weight (PTNBW) singleton deliveries.
Data collection: Interview and clinical examination
The interview was conducted by a single person using structured questionnaire to gather information on oral health practices, maternal health status and the related general traditional risk factors for PTLBW infant deliveries, while the interviewer was masked (single-blind) to case-control status. In relation to alcohol consumption, tobacco smoking, exposure to environmental tobacco smoke (ETS) also known as passive or second-hand tobacco smoking the response was set as "yes" or "no" without inquiring for more details regarding quantities, frequency and duration of exposure (e.g. total hours exposed to ETS). A different person (apart from the one conducting interviews) collected specific information on case-control status from the patient's hospital records.
Clinical intraoral examination involved all teeth including third molars, whereby periodontal examination was done on six sites per tooth (buccal-mesial, mid-buccal, buccal-distal, lingual-mesial, mid-lingual and lingual-distal). The tip of the Williams periodontal probe was used to collect plaque from the tooth surface and recorded accordingly as either "present" or "absent". Presence of dental calculus, gingival bleeding on gentle probing, and periodontal probing pocket depth (PPD) was scored separately using the Community Periodontal Index (CPI) clinical probe that has a bulb at the tip (0.5 mm in diameter) with graduations at 3.5 mm, 5.5 mm, 8.5 mm and 11.5 mm . The hierarchical CPI scoring system was not used in this study. A dichotomous scoring system (present or absent) was used for dental calculus and gingival bleeding separately. For periodontal probing pocket depth, categorization of the PPD was done to encompass four different groups (0–3.5 mm, > 3.5 to ≤ 5.5 mm, > 5.5 to ≤ 8.5 mm, and > 8.5 to ≤ 11.5 mm). Gingival recession was scored in millimeters using the "Williams" periodontal probe that had graduations at 1, 2, 3, 5, 7, 8, 9 and 10 mm .
Loss of attachment (LA) on tooth surface was computed from the collected clinical data using an indirect approach, "extrapolated loss of attachment (ELA)" as follows: Presence of gingival recession of 1+mm accompanied by PPD 4+mm, operationally was considered as having loss of attachment extending to 5+mm. The operational criteria selected for the clinical definition of study participants who positively and unequivocally exhibited "periodontal disease diagnosis threshold (PDT)" was the presence of both periodontal probing pocket depth of 4+mm in four sites or more and gingival bleeding on gentle probing in 30% or more of the sites that were examined.
Dentition status was assessed according to the methods recommended by the World Health Organization (WHO) . In addition, very advanced carious lesions with carious pulpal exposure were scored separately as follows: Carious pulpal exposure (CPE) – when a tooth had an open deep dentinal carious lesion, with long standing provoked or spontaneous pain that did not subside even after the removal of the irritating factors; signs and symptoms of pulp involvement such as periapical abscess, tenderness to vertical percussion, gumboil, extra-oral facial swelling in close proximity to a carious tooth; carious tooth with polyp; and carious root stumps. Due to constraints in resources, no periapical radiological investigation was performed to confirm the clinical diagnosis.
Calibration and reproducibility
One calibrated examiner against a "gold standard" examiner performed all the clinical examinations in a single blind to case-control status of the study participants, as the examiner was masked of obstetric data. No calibration was done for gingival bleeding and microbial plaque because for these conditions, it is practically difficult to reach a good reproducible level. During data collection phase, duplicate clinical examinations were done to about 10% (thirty six) of the study participants. Using Kappa test, reproducibility for 6,912 examined sites (kappa value ± standard error, and the 95% confidence interval) for PPD was 0.92 ± 0.01, 0.91 – 0.94), gingival recession (0.78 ± 0.01, 0.76 – 0.81); and 1,152 teeth assessed for scoring dentition status (0.89 ± 0.02, 0.84 – 0.93) and carious pulpal exposure (0.88 ± 0.02, 0.83 – 0.92) whereby in all conditions assessed, the level of significance test was P < 0.001.
The Muhimbili University College of Health Sciences (MUCHS) Research and Publication Committee approved the protocol of this study. The Ethical Committee at MUCHS granted ethical clearance. A written consent was obtained from each study participant before commencement of the interview and clinical examination.
Data management and statistical analysis
Statistical package and level of significance
Data were entered into a personal computer and analyzed using the statistical package for social sciences (SPSS) 10.0 for Windows. Reproducibility calculations are presented as Kappa values with significance levels and the 95% confidence interval (CI). The results are presented as prevalence among cases and controls (for categorical variables) together with Chi-Square test, Odds ratio (OR) and the 95% CI; whereas for continuous variables, the mean values are presented accordingly. For statistical tests, 2-sided type 1 error probability < 0.05 was considered as the level of significance.
Transformation of data
During analysis, the number of active tobacco smokers was negligible (three participants only) while there were hundreds of passive or second-hand smokers. These two groups were therefore combined to form one group of "passive smokers" versus those not exposed to tobacco smoke. Also the study participants who qualified for the "ELA" description were categorized as one group against all others with no "ELA" characteristics.
For the purpose of describing the "extent and severity of periodontal disease" in this study population, some principles as described by Carlos and Coworkers for "extent and severity index" were employed . In this study, the "extent of periodontal disease" was expressed in percentages: the number of sites affected with PPD 4 mm or more (4+mm) was the numerator and total number of examined sites was the denominator. The "severity" score for periodontal disease was computed as the mean number of sites with probing periodontal pocket depths 6 mm or more (6+mm) that is the sum of all sites with PPD 6+mm divided by the number of study participants examined.
The "periodontal disease diagnosis threshold (PDT)" categorization was computed as follows: First, the total number of sites in the study participants that exhibited gingival bleeding on gentle probing was counted and those that were ≥ 30% were categorized as one group from those which had < 30% of the sites affected. Secondly, the total number of sites with periodontal pockets 4+mm was counted and the study participants that had four or more sites were categorized as one group from those who had less or equal to three affected sites. Thirdly, the PDT criteria were worked out based on approaches used for PDT elsewhere and the classification of periodontal disease by the American Academy of Periodontology [8, 9, 32]. The PDT was scored as "yes" if the study participant had exhibited both gingival bleeding in ≥ 30% of the examined sites and probing periodontal pocket depth 4+mm in four or more sites, and "no" for those with lower values than the above set criteria.
In order to find out whether the relationship between periodontal disease threshold and preterm low birth weight is confounded, stratification in the cross-tabulation was done for PDT, PTLBW and other potential confounding factors as recommended by Ylöstalo and Knuuttila . Therefore, a stratum-specific relative risk (RR) estimates resulting from the 2 × 2 stratified cross-tabulations for the association between periodontal disease threshold, preterm low birth weight infant delivery and other potential confounders particularly age, hypertension, number of infant deliveries (parity), history of previous LBW deliveries, heavy duty, cigarette smoking, alcohol consumption and late onset of prenatal care were generated. The potential confounding factors were categorized as either "present" or "absent" (dichotomized). Since the total number for PDT was only 21 study participants, stratification resulted into smaller numbers equal to less than five in a cell of the 2 × 2 table thus necessitating the use of Fisher's Exact test rather than the Chi-Square test.
The demographic and maternal characteristics included in the univariate analysis were age, education, marital status, tobacco smoking, exposure to environmental tobacco smoke (ETS), alcohol consumption, heavy duty during pregnancy, prenatal care, place of delivery, and parity. Also specific oral health conditions in particular microbial plaque, dental calculus, gingival bleeding (on tooth brushing and as well on gentle probing), periodontal pockets, gingival recession, decayed-missing-filled teeth (independently and as a cumulative index – DMFT) were all incorporated in the univariate analysis and presented accordingly.
Multivariate logistic regression model
Multivariate logistic regression model was developed to examine the association between maternal oral health status in particular specific periodontal conditions (periodontal pockets, gingival bleeding, gingival recession, calculus), combined periodontal conditions "periodontal disease threshold", and dentition status (decay-missing-filled teeth and open pulp pulpitis) as risk factors for preterm low birth weight infant delivery. This was done with the understanding that PTLBW is multi-factorial in nature involving demographic, genetic, nutritional, obstetric, antenatal care, oral hygiene, professional dental care, dental plaque, poor oral health, periodontal disease, infection, maternal morbidity and toxic exposure  whereby the interaction is not strictly in a hierarchical manner as described by Victoria and Coworkers  and presented by Bassani and Coworkers . In the multi-factorial conceptualization, demographic and social factors of importance were age, education, parity, prenatal care, alcohol consumption, tobacco smoking and heavy duty during pregnancy. Environmental tobacco smoke (ETS) was considered as an environmental risk factor for PTLBW. The logistic regression models used were the "Forward stepwise-conditional" based on the likelihood ratio criterion (pin = 0.05, pout = 0.10) and the "Enter" methods with both "continuous" and "categorical" variables in the model, and in a different phase, a model with "categorical" variables only was developed, accordingly.
The first model analysis included all study participants, both the primi- and multiparous mothers (n = 373), cases (n = 150) and controls (n = 223) whereby the binary dependent variable was PTLBW (yes = PTLBW, no = TNBW) and the covariates were all those that were statistically significant or near the significant level in the univariate analysis. These factors included as covariates were age, marital status, high blood pressure, parity (number of deliveries), frequency of tooth brushing/day, missing teeth, tooth sites that had calculus and gingival bleeding.
In the second model, the analysis included all study participants (n = 373) and variables as in model-one (above), with an addition of new covariates that were considered biologically important based on literature review though not statistically significant in the univariate analysis (forced-entry model). These additional factors were malaria, tobacco smoking and exposure to environmental tobacco smoke (ETS), heavy duty during pregnancy, consumption of alcohol, late antenatal care, home delivery, open pulp pulpitis, periodontal disease threshold, and previous history of low birth weight. Calculus and gingival bleeding were excluded in this model because they are basically in one way or another considered within the periodontal disease threshold.
On the third attempt, the analysis dealt with only the primiparous mothers (n = 206) whereby cases and controls were (n = 92) and (n = 114), respectively. Except for parity that was excluded, the covariates were exactly the same as those included in model-one above.
Again, in this model, analysis dealt with only the primiparous mothers (n = 206), whereby all the covariates in model-two (above) were included, except parity and previous history of LBW, as the study participants were primiparous mothers.
The selected final model was the one that included variables that were statistically significant (P < 0.05) or near significant (P < 0.1) in the univariate analysis. At this stage, in the logistic regression model, the "Enter method" using dichotomized variables based on "yes" or "no" response (e.g. exposure to ETS) and median split values (e.g. for calculus and gingival bleeding) was selected instead of a "combination of both the continuous and dichotomized" variables. There was no "continuous variable model" for the whole data set, due to the fact that the type of data collected from the study population did not improvise for such transformation.