healthmedicinet

Source of data

The Hong Kong Cancer Registry is an official member of the International Association
of Cancer Registries 13]. Detailed data of cancer incidence and mortality are documented in its official website
covering the time frame from 1983 to 2012 12]. The following cancers were included into the study as OCC: cancers of base of tongue
(ICD 10: C01), other and unspecified parts of tongue (C02), gum (C03), floor of mouth
(C04), palate (C05), and other and unspecified parts of mouth (C06).

Statistical analysis

The ASRs of each gender for each year between 1986 and 2010 were calculated by using
the direct method for age-adjusting, according to the World Health Organization World
Standard Population 2000. Cases over 20 years of age were categorized into three age
groups by gender: Group 1: 20 to 44 years of age, Group 2: 45 to 64 years of age,
and Group 3: aged 65 or above. For each age group, the annual age-specific incidence
rates between 1986 and 2010 were calculated. Groupings and calculation of age-standardized
incidence rates and age-specific incidence rates were performed directly in the website
of the Hong Kong Cancer Registry with its own Cancer Statistics Query System 12].

To estimate the secular trends of OCC incidence in Hong Kong, the annual percentage
change of incidence rates was calculated for the entire period of observation, 1986
to 2010, both for each age group and also the entire population of each gender. The
annual percentage change estimation was based on the assumption that the incidence
rates changed constantly and linearly on the log-scale 14]. Besides, a joinpoint analysis was conducted to investigate the chronological point
in time of the trend change in a linear slope 15]. In this joinpoint analysis, significant changing points of the incidence trends
and annual percentage change values for each segmented trend line were computed. For
all calculations, a P-value of less than 0.05 was considered to be statistically significant.
All calculations of the annual percentage change and joinpoint analyses were performed
with the ‘Joinpoint’ software version 3.5.4 provided by the US National Cancer Institute
on a Windows XP Home Edition (National Cancer Institue, Bethesda, MD).

An age, period, and cohort (APC) model was used to analyze the influences of chronological
age, population’s birth cohort, and time period of the diagnosis 16]. With respect to the chronological age, the population was divided into ten groups;
nine quinquennial age groups from 40 to 84 years of age, and a group of 85 years of
age or above. Regarding the time period of diagnosis, the entire 25 year observation
period from 1986 to 2010 was split into five intervals, each encompassing a period
of 5 years. Fourteen overlapping birth cohort periods were set between 1896 and 1965
at 10 year intervals, starting with the cohort from 1896 to 1905, ending with the
one from 1961 to 1970. Related to the assumption that OCC incidence rates follow the
Poisson distribution, seven statistical models in log-linear scale were established
containing the following effects; age-only (A), period-only (P), cohort-only (C),
age-period (AP), age-cohort (AC), and full APC. Akaike’s Information Criteria (AIC)
of each model was confirmed to estimate their reliability of the models.

Due to linear dependences between age, period and cohort effects (Cohort?=?Period
– Age), it is well understood that the concomitant estimation of these three effects
is impossible. This fundamental “identification problem” of the APC model estimation
was tackled with the following mathematical measures:

1) AIC calculation of A, P, and C models.

2) As the P model demonstrated the least goodness of fit depicting nearly a simple
linear function for both genders, an alternative P model was calculated, including
fewer parameters and a better goodness of fit. The P model where P was replaced by
1) a constant (y?=?a, a is constant) was named “P-constant” model; 2) a linear function (y?=?ax?+?b, a and b are constants) was termed “P-linear” model; and 3) a quadratic function (y?=?ax^2?+?bx?+?c, a,b and c are constants) was called “P-quadratic” model.

3) AICs of the above three alternative period models were calculated to find out the
best fitted alternative model for each gender group.

4) Better matching period models were extrapolated to AP and APC models in order to
achieve better fitting and to overcome the “identification problem” of a full APC
models.

All calculations related to age, period, and cohort analysis were performed on Macintosh
OSX version 10.8.1 with ‘JMP 10 (for Macintosh)’ software version 10.0.0. (SAS Institute
Inc., Cary, NC) 17].