We conducted separate APC model calculations for the periods of 1981 to 1995 and 1996 to 2010 to verify the hypothesis that the cervical screening program influenced the mortality trends after 1995. The results not only supported the positive effects of the screening program, but also showed unexpectedly patterns. First, the ASMR results and the period effects of the APC model both indicated that cervical cancer mortality trends were reduced after the screening program involving Papanicolaou test was implemented in 1995. Additionally, many previous studies have reported on the efficacy of the screening program in Taiwan [15–18]. Second, the effects of the screening program were not observed among elderly women.
The differential period ratios for cervical cancer show a decline in mortality rates for people between 30 and 79 years of age after the implementation of the screening program. By contrast, the risk of mortality caused by the protectiveness of the screening program was 1.4-fold higher than that during the periods before the screening program was implemented among patients between 80 and 84 years of age. These contradictory patterns may indicate a problem. Additionally, the screening rates among elderly women in Taiwan may be lower, preventing early diagnosis and treatment, thereby increasing their mortality rate and reducing their survival rate. However, the answer to this issue was found in the statistical data obtained from the Bureau of Health Promotion of the Department of Health. The data indicated that the screening rates were 55% to 65% for women aged between 30 and 39, 50% to 60% for women aged between 40 and 49, 45% to 55% for women aged between 50 and 59, 40% to 50% for women aged between 60 and 69, and 25% to 32% for women aged 70 or more. Thus, the screening rates were lower among elderly women.
The cervical cancer mortality rate for older age groups (70+) exhibited an increasing trend before 1995, but not among younger women. i.e. the mortality risks for women aged 70 years or more increased annually before 1995; whereas no significant change in mortality trends was observed among women under 70. Present studies cannot provide a suitable explanation for this finding. The reason for these results may be a particular risk factor directly or indirectly related to age that remains unknown. However, we still propose some possible reasons causing this phenomenon for subsequent study validation. Regarding the phenomenon of multiple births among women of the post-war generation, a number of previous studies have suggested that pregnancy at an excessively early age and excessive pregnancies may be risk factors for cervical cancer [19–22]. Taiwan experienced a post-World War II baby boom. Beginning in 1947, the birth rate increased from 38.31 to 49.97 by 1951. Consequently, we can infer that women who were 70 to 84 years of age between 1980 and 1995 were of a reproductive or childbearing age (i.e., 20 to 35 years of age) from approximately 1930 to 1960. This includes all women at childbearing age during the post-war baby boom.
However, the results of numerous previous studies also support the limited effectiveness of cervical cancer screening among elderly women [23–27]; one study even contended that screening elderly women was inefficient and should be terminated for women aged 65 or older who have a history of regular negative smears . Considering this argument, we contend that if screening is genuinely ineffective among elderly women, then the mortality rate should remain unaffected. However, this is completely unreasonable in Taiwan. The pattern of age-specific mortality rates for each age group shown in Figure 4 exhibited a declining trend among elderly women. The results also indicated that the effectiveness of the screening program may not immediately affect women between 80 and 84 years of age in Taiwan because of treatment delays during the initial screening program implementation stage. Previous studies related to the effectiveness of screening have considered the problem of a lead time bias. However, the idea that reducing incidence was the main screening effect that could successfully reduce mortality rates was based on the notion that no deaths would be caused by a disease if the disease did not occur. However, the reduced mortality rate caused by screening is not entirely because of a lower or reduced incidence rate. Especially when cancer prognosis results are good, implementing screening may increase the incidence rate, but it can reduce mortality rates through early treatment. For example, in Taiwan, a few years after the screening program was implemented in 1995, the incidence of cervical cancer increased significantly; however, the corresponding mortality rate declined following the implementation of the screening program. Thus, the reduced mortality rate for cervical cancer cannot be mainly attributed to a lower incidence; instead, it was also affected by whether the Papanicolaou test provided good prognosis results, thereby increasing the patient survival rate.
In 2006, Lloroa  contended that the low cervical cancer mortality rates were not only the result of the Papanicolaou test or other screening techniques and treatment reforms, but could also be attributed to misclassifications on death certificates. The study showed that numerous uterine cancer cases were classified as “site unspecified.” According to Cuzick and Beral [29, 30], uterine cancers can be classified into three types, that is, cervical cancer, corpus uteri cancer, and cancer of the uterus at an unspecified site (UNOS). However, Levi  believed that decrease in cervical cancer and increase in UNOS may be caused by the revision of the International Classification of Diseases (ICD). Therefore, the proportion of UNOS declines should be considered to examine changes in cervical cancer mortality. Furthermore, to obtain an accurate calculation for long-term analysis, Lloroa suggested that the proportion of UNOS and uterine cancers should combined to determine cervical cancer mortality rates .
However, this assumption and concept do not need to be considered in this study. Furthermore, the above conditions indicate that the study design included analysis of long-term trends and more than one version of ICD codes. In the studies conducted by Lloroa and Levi, the periods analyzed were 1955 to 1995 and 1960 to 1998, and they included 4 versions (ICD-7th to 10th) and 5 versions (ICD-6th to 10th) of ICD codes. To the best of our knowledge, the different versions of ICD codes were developed through diagnosis and treatment. Thus, we can infer that the old version may be the primary cause of the assumption. Based on this condition, we calculated the cervical cancer mortality trends for 1981 to 2010, but we only included two versions of ICD codes. However, the death registration system in Taiwan cited ICD-9th for death certificates for 1981 to 2008, and ICD-10th for 2008 to the present. Therefore, the effects of various versions of ICD codes are the lowest in this study.
The accuracy of cause-of-death coding in Taiwan appears to vary according to the type of disease. Disagreements between the reviewer and the original coder included disagreements regarding the nomenclature, inappropriate judgments of cause relationships, and incorrect interpretation of the International Rule for Selecting the Underlying Cause of Death, and Modifying the Selected Underlying Cause of Death. The cause-of-death data used in this study was analyzed in 1994 and published in 2000. Therefore, the early cause-of-death data may influence the accuracy of the study results; this is the inevitable study limitation.
The quality indicators included the percentage of morphologically verified cases (MV%). The percentage of death-certificate-only cases (DCO%) between 1980 and 2009 show steady improvements in the quality of Taiwan Cancer Registry. The MV% increased from 82.4% between 1980 and 1984 to 89.05% between 2005 and 2009. The DCO% declined from 28.8% between 1985 and 1989 to 1.4% between 2005 and 2009.