Assessment of Registry-Based Surveillance Statistics Used for Cancer Control in the Dindigul District in South India

Abstract

Background:

Cancer incidence rates from the Dindigul district were lower by 50% than Chennai in Tamil Nadu for most cancers. This study describes the cancer surveillance statistics and provides an assessment of missing cases from routine registration in the Dindigul Ambilikkai Cancer Registry (DACR), covering a predominantly rural population in the Dindigul district.

Method:

A total of 21,214 incident cancers in the DACR during 2003–2017 were examined for this study. Cancer registration was carried out by active case-finding following standard international norms. A total of 12,541 incident cancers registered during 2003–2012 and followed through 2014 were used to estimate survival. Data on follow-up were obtained through a mixture of active and passive methods. Survival probability was estimated by actuarial methods. A random survey carried out independently was used to assess the quality of case ascertainment.

Results:

The age-standardized rate (ASR) per 100,000 population was higher among women (76.2) than men (61) in 2013–2017, with both sexes reporting a 17% increase compared to 2003–2007. The most common cancers were cervix (ASR,18.5) and female breast (ASR,17.1), with percentage changes of –19% and +46.1%, respectively. Lung cancer (ASR, 5.5) was top among men with an increasing trend (+57.1%). The percent change in ASR of mouth cancer showed opposite trends among men (+24.3%) and women (– 21.4%). The ASR of colorectal cancers almost doubled among men between 2003–2007 and 2013–2017 (3.9; +94.7%). The 5- and 10-year absolute survival for all cancers were 31% and 20%, respectively. Out of 365 incident cancers that occurred during 2003–2010 in the surveyed areas, 310 (84.9%) were already registered in the DACR, while 55 were newly identified from the survey (15.1%). Inadequate coverage of sources outside the Dindigul district was significant (P = .002), with the highest number of missed cases from hospitals under nongovernment sectors (58.3%). Underascertainment was higher among cancer patients living in hilly regions (60%) and border areas (47.4%) than in core regions (P = .05).

Conclusion:

Because of an enacted government order making cancer a notifiable disease, the registry-based cancer surveillance could be extended, covering a population of 80 million in a cost-effective manner with enhanced coverage and systematic evaluation of cancer-screening programs.

Introduction

Cancer surveillance using a population-based cancer registry (PBCR) is essential for cancer control, with incidence, mortality, and survival as cornerstones. Though the number of PBCRs providing data from regions in the low- and middle-income countries (LMIC) is growing, there is still inadequate coverage of newer populations.¹ Surveillance through systematic registration of new cancers is currently happening in India through the network of 33 PBCRs under the National Cancer Registry Programme of the Indian Council of Medical Research² and 10 PBCRs under other programs in India. Still, cancer registries covering rural populations are sparse in India, making the estimation of disease burden for the entire country cumbersome. The Dindigul Ambilikkai Cancer Registry (DACR), a PBCR covering a predominantly rural population since 2003, was organized by the Cancer Institute (WIA) of Chennai in Tamil Nadu, India, and the Christian Fellowship Community Health Centre in Ambilikkai, Dindigul, where field operations are based. The Dindigul district is one of 31 districts in Tamil Nadu and has a population of 1.8 million. The incidence rates of the most common cancers in the rural Dindigul district were almost half of the corresponding ones observed in the metropolitan population in Chennai in the same state of Tamil Nadu.³ An assessment of completeness of coverage of cancer cases is essential to understanding whether the low incidence rates reported in rural populations is real. This study describes cancer incidence and survival statistics produced using data from the DACR. Additionally, missing cancer cases identified through a household survey independent of routine cancer registration in the DACR is examined to assess data completeness and consider potential solutions for data quality improvement.

Methods

Population-Based Cancer Registry (PBCR)

The source of data for this study is 21,214 incident cancers registered in the DACR during 2003–2017. Cancer was not a notifiable disease in the registry area during the study period. Hence, the registration of incident cancers in the DACR was essentially carried out by active case-finding. The registry staff were trained in cancer terminology to identify cancer diagnoses from case records and pathology registers maintained in health care organizations. The potential sources of data for the DACR included hospitals in government and nongovernment sectors, stand-alone pathology laboratories, imaging centers, hospices, and the Tamil Nadu cervix and breast cancer screening programs. A standard form was used for data abstraction, comprising mandatory variables (source of registration, patient name, age at diagnosis, sex, residence area, incidence date, most valid basis of diagnosis, primary site of cancer) and optional variables (father/spouse/relative names, literacy status, clinical extent of disease, previous or current cancer directed treatment received with hospital details). Data on cancer mortality were abstracted from various vital statistics offices. Disease coding was done using the International Classification of Diseases for Oncology, 3rd Edition (ICD-O-3)⁴ and the International Classification of Diseases, Tenth Revision (ICD-10),⁵ adhering to international norms. An in-house-developed software tool was used to collect data. In the DACR, efforts were routinely made to contact all newly registered cancer patients by visiting their homes or by making telephone inquiries on an annual basis for completeness of details in the registry form.³ Duplicate notifications registered in the same and previous years in the DACR were meticulously eliminated using MatchPro⁶ and in-house programs. These listed the possible duplicate cases in pairs based on the similarity level of patient identification and disease details. The data were updated in the original as appropriate. Quality checks were performed on data validity and consistency of essential variables using in-house programs and other tools.

Age-standardized rates (ASRs) were computed as a summary figure of cancer incidence for comparison between different populations by multiplying the observed 5-year age-specific rates with respective weights given in Segi's world standard population.¹ Cancer incidence data of the first and the last 5-year periods were compared to elicit the percent change. Census norms were followed for urban or rural classifications.⁷ All incidence rates and cumulative lifetime risks of acquiring cancer were computed following the standard norms.¹

Overall Survival

A total of 12,541 incident invasive cancers registered during 2003–2012 formed the basis for the survival study. Cases registered based on a death certificate only with zero survival time (n = 668) were excluded, leaving 11,873 for analysis. Line-list data on cancer mortality from vital statistics offices were linked with the DACR registry data-base. Mortality data that matched with the incident case were updated in the DACR database. Data on the follow-up of the remaining incident cases without mortality data were obtained by active methods: inquiry by home visit/telephone, repeated scrutiny of the case record in source hospitals, and perusal of household registers maintained by public health workers in villages. The studied endpoint was death (irrespective of the cause). The survival duration was calculated as the time elapsed between the incidence date and date of death or date last known alive as of December 31, 2014. Absolute or overall survival probability was estimated for all ages together and by 5-year incidence period, age group, sex, and clinical extent of disease using actuarial method.⁸ Differences between survival curves for statistical significance were computed using a log-rank test.⁹

Household Survey

A household survey targeting 5% of the 2 million total population in the rural Dindigul district was carried out in 2011–2012 as a method of independent ascertainment of cancer cases to estimate the extent of missing new cancers from registration in the DACR. A proportional allocation of persons adhering to an urban–rural ratio in the population was performed to determine the sample size from each of the 7 subdistricts. The list of villages and towns in the district with an estimated population was used as the sampling frame. Villages and town localities were chosen by simple random sampling from each subdistrict, keeping in mind the allotted sample size. Special care was taken to include areas bordering the district and those from which no cancer case has been registered so far in the DACR to improve the representativeness. All inhabitants in the chosen villages/town localities were enumerated in full for inclusion in the survey. A total of 97,632 people in 19,333 residential households from 41 villages/town localities formed the survey material. A study questionnaire was devised, incorporating data items on family members in each household and history of cancer or death in the family since 2003 until the year 2010, with necessary identity information. This questionnaire was administered by trained health workers in the area unrelated to registry operations. Information was obtained from any adult about themselves and others in the household.

Results

Data quality indices for cancer registration in the DACR during the latest period of 2013–2017, consisting of 8,662 cases, are provided in Table 1. Data on person identifiers were adequate. Every patient's full name was known with no missing values for factors like stated age at diagnosis, sex, exact date of first diagnosis, and village/town name of the patient's primary residence. Data on cancer validity measured by histological verification of diagnosis was 78%, while approximately 9% of the cancer cases were of uncertain primary, and 9% were registered based on death certificate information only.

Table 1.

Data Quality Indices of Cancer Registration in Dindigul District, 2013–2017

Data quality indices	No.	%
Number of data sources	214	100.0
Located within Dindigul district	85	39.7
Located outside Dindigul district (number of districts)	129(24)	60.3
Person identity
With patient name in full	8,662	100.0
With at least 1 relative name (parent, spouse, offspring, etc)	6,036	89.0
With at least village/town locality name	8,662	100.0
Age at diagnosis	8,662	100.0
Sex	8,662	100.0
Date of first diagnosis	8,662	100.0
Histologically verified diagnosis	6,733	77.7
Primary site of cancer uncertain (ICD-10: C76-80)	784	9.1
Death certificate notification	808	9.3

The DACR incidence trends between 2003–2007 and 2013–2017 were upwards for most parameters (Table 2). The age-standardized rates of all cancers together were increasing, with positive percent changes both among men (17.5%) and women (17.4%). The percent change in age-specific rates representing children (aged 0–19 years) was higher among girls (38.3%) than boys (15.3%). Age-specific incidence rates among matured adults and geriatric patients between the two 5-year periods showed an increasing trend among both sexes (ages 40–64 years: men, 10.1%; women, 11.5%; ages ≥65 years: men, 34.7%; women, 57.5%). However, among the young adult group (ages 20–39 years), the rates were almost static among men (+2.9%) but fell among women (–12%). Trends and cancer incidence rates in both 5-year periods were higher in rural versus urban areas for both sexes.

Table 2.

Incidence Statistics of All Cancers Together in Dindigul District Among Men and Women: 2003–2007 vs 2013–2017

Measure of incidence	2003–2007		2013–2017		% Change
	Men	Women	Men	Women	Men	Women
Average annual number of new cancers	516	677	750	982	45.3	45.1
Men:women	1:1.3		1:1.3
Crude incidence rate per 100,000 population	50.8	67.2	66.1	86.4	30.1	28.6
ASR per 100,000	51.9	64.9	61	76.2	17.5	17.4
Cumulative risk % (0-74 y)	1 in 17	1 in 15	1 in 15	1 in 12
Age-specific rate by age group (y)
0-19	5.9	4.7	6.8	6.5	15.3	38.3
20-39	17.2	33.4	17.7	29.4	2.9	-12.0
40-64	112.4	172.7	123.7	192.6	10.1	11.5
≥65	229.3	155	308.8	244.1	34.7	57.5
Urban - ASR	33.9	62.1	45.9	79.1	35.4	27.4
Rural - ASR	58.5	78.5	78.5	102.5	34.2	30.6

ASR, age-standardized rate, a summary figure of cancer incidence for comparison between different populations by multiplying the observed 5-year age specific rates with respective weights given in Segi's world standard population.

The common cancer incidence rates and trends comparing 2003–2007 and 2013–2017 are depicted in Figure 1, where ASRs for both diagnosis-year groups stratified by sex are provided. Cervix (ASR, 18.5) and female breast (ASR, 17.1) remained the top-ranking cancer sites among women, although the percent change in ASR of the cervix (–19%) and breast (+46.1%) showed opposite trends. Lung cancer (ASR, 5.5) was the most common among men and showed an increase (57.1%) while stomach cancer showed a minimal decrease (–5.2%). Percent change in mouth cancer incidence showed opposite trends among men (ASR, 4.3; +24.3%) and women (ASR, 2.1; –21.4%). ASR of colorectal cancers almost doubled (+94.7%) among men (ASR, 3.9) while the percent change was 41.7% among women (ASR, 2.0). Cancer of the body uterus (corpus uteri) emerged within the top 5 in 2013–2017.

Figure 1.

Age-Standardized Incidence Rates for Common Cancers by Sex, Dindigul district: 2003–2007 vs 2013–2017

Figure 2 displays the 5-year absolute survival percent from common cancers registered in the DACR during 2003–2012. The highest survival reported was for thyroid (69%) followed by body uterus (59%) and female breast (50%). Stomach (10%), lung (10%), and esophageal (9%) cancers experienced the least survival. The 5-year and 10-year absolute survival (not displayed) for all cancers were 31% and 20%, respectively.

Figure 2.

Five-Year Absolute Survival Percent of Common Cancers in Dindigul District: 2003–2012 Cases Followed Until the End of Year 2014

The survival experience of all cancers diagnosed from 2003–2012 are displayed in Figure 3. The 5-year survival of cases registered in 2003–2007 (30%) was almost the same as in 2008–2012 (29%; Figure 3a). There was a decreasing trend in survival by increasing age groups (Figure 3b): 5-year survival from cancer between those aged 0–19 years and 20–39 years at diagnosis did not differ significantly (P = .985), while in comparison to patients aged 0–19 years, the survival experience in those aged 40–64 years and ≥65 years were both lower and statistically significant (P < .001). Women had significantly better survival than men (P < .001; Figure 3c). Figure 3d displays the overall survival by clinical extent of disease. Localized cancers had the highest survival, while those presenting with distant metastasis had the poorest survival. Survival of cancers with regional spread of disease and not applicable or unknown categories were in between, with survival rates of each of the categories being significantly different from the rest.

Figure 3.

Absolute Survival in Dindigul District: 2003–2012 Cases Followed Until the End of Year 2014—All Cancers Together

Case ascertainment by DACR registration and the independent household survey in the randomly chosen areas within the Dindigul district revealed a total of 365 incident cancers in the district during 2003–2010 (Table 3). Out of these, 310 (84.9%) were already registered in the DACR, representing the completeness of coverage by the DACR. Concordant cases representing new cancers that were both registered in the DACR as well as were identified in the survey comprised 111 (30.4%), while new cancers missed from registration in the DACR but were identified only from the survey accounted for 55 (15.1%).

Table 3.

Completeness of Casefinding in Dindigul Ambilikkai Cancer Registry (DACR) Assessed by Independent Ascertainment by Household Survey, 2003–2010

Eligible cancer cases	No.	%
Concordant cases: Cases both registered in DACR and identified in survey (A)	111	30.4
Discordant cases: Cases registered in DACR but not identified in survey (B Cases identified in survey but not registered in DACR (C)	199 55	54.5 15.1
Total cancer cases from surveyed areas, either registered only in DACR or identified only in survey or in both (A+B+C)	365	100.0
Completeness (%) of reporting of cancer cases by DACR	A+B out of A+B+C (84.9%)
Completeness (%) of reporting of cancer cases by survey	A+C out of A+B+C (45.5%)
Cancer cases missed from registration in DACR	C out of A+B+C (15.1%)

Table 4 provides factors associated with cancer cases being missing from the DACR during the 8-year period of 2003–2010 among the 166 total cancers identified in the household survey. The number of missing cases from the DACR were more frequent in the latter period of 2007–2010 (38.5%) compared to 2003–2007 (25.7%). They were also more frequent among women (36.5%) compared to men (28.6%), and among children or young adults than those aged ≥35 years, although none of these were statistically significant, displaying random variation. Inadequate coverage of sources outside the Dindigul district emerged as statistically significant (P = .002). The highest number of missing cancer cases were among those that were diagnosed or treated in hospitals under private or nongovernment sectors (58.3%) compared to the least among hospitals under the government sector (23.1%) located outside of the Dindigul district. Cancer patients living in hilly regions (60%) and in the border areas (47.4%) were more likely to be missing from the DACR than those residing in core regions of the district (29.2%) (P = .05). Patients from distant areas or without direct access to the base institution were more likely to be missing from the DACR (33.6%) than those living in proximity.

Table 4.

Factors Associated with Cancer Cases Missing from Registration in Dindigul Ambilikkai Cancer Registry (DACR) Among Those Identified from Survey (N = 166)

Factor	Cancer cases identified from survey
	Already registered in DACR (n = 111)		Missed from registration in DACR (n = 55)		P value
	No.	%	No.	%
Year of diagnosis					.08
2003–2006	52	74.3	18	25.7
2007–2010	59	61.5	37	38.5
Sex					.28
Male	50	71.4	20	28.6
Female	61	63.5	35	36.5
Age group (y)					.62
0–19	3	60.0	2	40.0
20–39	5	50.0	5	50.0
40–64	73	69.5	32	30.5
≥65	30	65.2	16	34.8
Resident status					.973
Urban	81	66.9	40	33.1
Rural	30	66.7	15	33.3
Type of source of registration					.002
Private hospital outside Dindigul district	15	41.7	21	58.3
Private hospital inside Dindigul district	61	74.4	21	25.6
Government hospital outside Dindigul district	30	76.9	9	23.1
Government hospital inside Dindigul district	5	55.5	4	44.5
Location of residence					.05
Hilly regions	4	40.0	6	60.0
Core region	97	70.8	40	29.2
District border	10	52.6	9	47.4
Distance from registry base institution					.034
Distant areas or areas not having direct access to the base institution (≥30 km)	101	66.4	51	33.6
Areas in proximity or having direct access to the base institution (<30 km)	10	71.4	4	28.6

Discussion

The distribution and development of cancer health services are generally lopsided, favoring urban versus rural settings in most low- and middle-income countries. In India, the nongovernment or private sector provides health services on out-of-pocket payments with significant patronage from the sick. During the study period, in 32 districts of Tamil Nadu, 16 had no radiotherapy (RT) facility at all, and in 9 districts, RT was available from the nongovernment sector only. In the Dindigul district, RT was available only in the base institution under the nongovernment sector, while 2 of the more developed neighboring districts had more than one, encompassing the government and nongovernment sectors. Also, histopathological and imaging services are not well developed in the Dindigul district. Hence, outflow to other districts for cancer care is anticipated, making cancer registration very challenging. This was addressed by enrolling 69 cooperating data sources from 10 districts other than Dindigul during the initial years of DACR operations.³

Data quality indices pertaining to eliminating duplicate notifications by probabilistic matching were adequate, with excellent availability of patient identity information in the DACR. The fact that DACR data have conformed to international standards stands as testimony to other aspects of data quality.¹⁰ Furthermore, the DACR frame-work was utilized in evaluating a field trial on cervix cancer intervention¹¹ and in predictions.¹² However, assessing the completeness of coverage by indirect means (through official mortality statistics based on a deficient system) will be incorrect, and capture–recapture methods rely entirely on existing data sources.

Evaluating the completeness of coverage by a PBCR through an independent case-ascertainment mechanism is the most desirable form of estimating the missing incident cancers from routine registration. A survey was conducted of randomly chosen households unrelated to registry operations for identifying cancer cases by trained health workers. We discovered that conducting a survey after a long lag time was not ideal for identifying cancers in this community, primarily due to reasons related to patient confidentiality and family stigma. However, this survey revealed a definite association with missing incident cancers from the district registry in predominantly rural settings through surrogate factors that implied a lack of development and accessibility to health services. Hence, a continuous expansion of data sources not restricted to the registry coverage area is viable and within the scope of any registry for improving coverage, thereby reducing missing new cancer cases. This was achieved by improving the number of sources of registration in the DACR from 24 districts to 129 districts outside of Dindigul after the conclusion of the survey analyses. Subsequently, the average annual number of new cancers registered in the DACR during 2013–2017 improved from 1,317 to 1,732 during 2008–2012, showing a 31.5% increase. This offset the missing proportion of cases in earlier years, including the fixed demographic effect. It also demonstrates that the lower cancer incidence reported in the DACR compared to predominantly urban districts or metropolitan cities in Tamil Nadu and the rest of India is real.^2,13

Although 15 years is short for any formal analysis of incidence trends, the average percent change in ASR provides valuable leads for cancer control planning. The steadily declining ASR of cervix cancer even in a rural district like Dindigul may in part be due to the effective field intervention trial¹¹ and other government-funded opportunistic screening programs. In the Dindigul district, more women in rural areas (18.5%) received screening tests compared to urban areas (16.5%).^14,15 The increase in breast cancer in the DACR probably signals the steady increase in urbanization happening along with associated lifestyle changes. The opposite trends in mouth cancer incidence by sex (increasing among men but decreasing among women) are perplexing. The National Family Health Survey (NFHS) revealed that the use of tobacco in any form was less in women (3%) than men (31.2%), and smokeless tobacco and alcohol use have risen, though tobacco smoking has declined among adult men in Tamil Nadu.¹⁶ It is only reasonable that tobacco use in any form has reduced with education among women but not among men, calling for revisiting the tobacco control strategy.¹⁵ The newer threat that is looming is the emergence of colorectal cancer, now within the top 5 among both sexes in a majority of urban and rural areas of Tamil Nadu, including Dindigul.¹³ The ASR for colorectal cancer is continuously increasing in the rest of India among both sexes.² With an estimate of 1,750 new cases being added every year in Tamil Nadu¹³ and the changing dietary practices among the younger generation, a fecal immunochemical test (FIT) based screening on established lines¹⁷ was planned and pilot-tested for integration with the existing public health services in Tamil Nadu.¹⁴ It is intriguing that the lung cancer incidence in rural Dindigul is half that in metropolitan Chennai,¹³ though smoking prevalence is almost the same in rural (33.1%) and urban (32.2%) areas,¹⁶ bringing to the fore the possible causative role of outdoor air pollution.

Population-based cancer survival has always remained a special study rather than a routine in most LMICs.¹⁸ In the DACR, inquiries by active methods on an annual basis for completeness of details has served well for optimizing follow-up data, especially on mortality, of all registered cases.³ The inverse association between the clinical extent of disease and survival augurs well for data consistency and reliability. The 5-year absolute survival rates for all common cancers in the DACR were either on par with or lesser than other rural areas in India and LMICs.¹⁹

Implications

The registry-based cancer surveillance has been expanded to cover the entire state of Tamil Nadu¹³ by enacting a government order making cancer a notifiable disease under a public health act²⁰ since 2018, setting a precedent for other states in India. This has enhanced the completeness of coverage in the DACR and provided a framework for systematic evaluation of cancer-screening programs¹⁴ throughout the state in a cost-effective manner. The surveillance statistics generated by the DACR have made a significant contribution in shaping the cancer policy for the Tamil Nadu state.²¹ The platform for evidence-based approach for meaningful cancer control measures and further research through analytical epidemiological and implementation studies has been laid for the future.