The incidence of lung cancer in Northern Ireland: 1991–1992

Diagnosis was based upon existing histological, radiological and clinical evidence which was considered by one of us (PW). Where histological or cytological evidence was lacking, the diagnosis was made by PW based upon available investigations and the clinical course of the illness. In this way, only cases of primary lung cancer (ICD 162, 9th revision) were included in the study.

An incident case was defined to be a person of any age resident within Northern Ireland who was newly diagnosed as having primary lung cancer (ICD 162, 9th revision) during the one year period 1/10/1991 to 30/9/1992. For example, as framed, the definition specifically excludes cases of mesothelioma (ICD 163) or secondary lung cancer.

Incident cases were identified during the study period using a multi-source notification scheme. Notification sources included: general practitioners, hospital physicians and surgeons, radiotherapists, pathology laboratories, the office of the Registrar General (NI) and the NI Cancer Notification Scheme.

A simple notification form was designed and circulated to hospital and community-based Physicians and Surgeons at regular intervals together with reminding letters. These forms were completed by the notifying Physicians or Surgeons and returned to the researcher. In addition the thoracic Surgeons provided copies of discharge letters for in-patient, out-patient procedures and admissions. Separate arrangements were made with the Radiotherapy department, Pathology laboratories, Registrar General’s Office and the NI Cancer Notification Scheme enabling them to notify cases routinely. Furthermore, PW regularly monitored the notifying agencies and inspected log-books, computerised listings and other records in individual hospitals and units to ensure that cases had not been inadvertently overlooked.

Notifications from individual agencies were computerised in the Department of Epidemiology and Public Health, The Queen’s University of Belfast, and linked using the patient’s name, sex and date of birth. A file of potentially incident cases of lung cancer was produced and up-dated regularly throughout the study period. Following notification, PW reviewed the hospital notes and general practitioner notes (if necessary). For example, even when a notification was based on a single source (e.g., death certificate data) the patient’s hospital notes were retrieved (via a master patient index) to determine whether the case fulfilled the study criteria, and, if so, to abstract other pertinent clinical data. This method was applied for all cases regardless of whether they were notified from single or multiple sources. In this way, the number of incident cases of primary lung cancer occurring in Northern Ireland during the study period, and the pattern of notification among the different sources, was ascertained and verified.

Population and mortality data for lung cancer (ICD 162) were obtained from the Registrars General of Northern Ireland [25, 26] (1991/1992) and Scotland (1991/1992), respectively. For Scotland, incidence data were obtained from the Scottish Cancer Registry, whereas the Office of National Statistics provided incidence and mortality data for England and Wales as did the National Cancer Registry in the Republic of Ireland (personal communications). In general, the data selected for comparison were the latest (population and matching incidence) available at the time of writing.

Standard statistical methods appropriate to the analysis of incidence rates were employed. In particular comparisons, it was found convenient to model the probability of incidence, \(\theta\), by the linear logistic function [27], whence the natural logarithm of the ‘odds’ on incidence is given by:where \(\alpha ,\beta ,\gamma\) and \(\delta\) are unknown parameters describing dependence on the three factors studied. The generalised linear interactive modelling package, GLIM [28], was used to fit various forms of model (1) to the incidence data considered later.

In addition, age-standardised rates were calculated using the direct method. The 5% level of statistical significance, which should be regarded as a nominal reporting level, was used in tests of statistical hypotheses and in the construction of confidence intervals (CIs).

During the study period 3814 notification forms were received relating to 1818 individual possibly incident cases of lung cancer. However, of these 1818, the following were excluded: 283 (15.6%) with the wrong diagnosis (for example, no proven malignancy, secondary lung cancer, lymphoma), 605 (33.3%) with a diagnosis outside the study period, 12 (0.7%) who were resident outside Northern Ireland and 18 (1.0%) who could not be traced. In all, 918 (50.5%) cases were excluded leaving 900 incident cases of primary lung cancer for analysis.

Overall, a total of 2739 notifications were received for these 900 incident cases (a ratio of 3:1). Table 1, which groups the sources into 4 major categories S01–S04, illustrates the value of the multi-source notification approach adopted in the study. For example, a single source study based on the NI Cancer Notification System (S04) would have identified only 361 cases, 40.1% of the true incidence, compared with a maximal 676 cases, 75.1%, using only GPs, hospital physicians and other specialists (SO1). No single source studied was sufficiently complete to have been used for serious epidemiological purposes. Moreover, a study employing 3 sources (S01 and S02 and S04) but ignoring death certification data (S03) would have missed 115 cases, 12.8% of the true incidence. Remarkably, only 137 cases (15.2%) were known to all 4 sources during the course of the study.

Of the 900 incident cases, 601 were male and 299 were female. The 1991 census for Northern Ireland enumerated the population as 1,577,834 persons (769,057 males and 808,777 females). Based on these figures, the incidence of lung cancer in Northern Ireland during the study year was 57.04 cases per 100,000 population (95% CI = 53.24, 60.84). The corresponding rates for males and females were 78.15 per 100,000 (95% CI = 71.78, 84.52) and 36.97 (95% CI = 32.69, 41.25), respectively. Overall, the male to female incidence ratio was 2.1:1.

Since mortality has been used consistently as a surrogate measure of incidence in Northern Ireland, it is interesting to quantify, for first time, the resulting bias. In 1991, 788 deaths from lung cancer (ICD 162) were reported in Northern Ireland, yielding a mortality rate of 49.94 per 100,000 population. This figure under-estimates the true incidence rate by 12.45% and is computed as 100 \(\times\) (57.04–49.94)/57.04). Since the mortality rate falls outside the 95% CI for the true incidence rate, we conclude that the incidence of lung cancer in NI is statistically significantly higher than the mortality. The 1991 sex-specific mortality rates were 73.60 and 27.44 for males and females and these figures underestimate the corresponding incidence rates by 5.82% and 25.78%, respectively. The underestimate for females is rather dramatic and highly statistically significant. A comparison with the mortality rates in 1992 reveals a similar pattern: the overall mortality rate under-estimates the true incidence rate by 16.02%; for males the underestimate is 16.68% and for females the underestimate is 14.93%. All of the mortality rates lie outside the 95%CIs for the corresponding true incidence rates.

Table 2 (a) shows the observed incidence rates (per 100,000) for Northern Ireland by age (10 year groups) and sex. Clear differences in incidence exist between sexes although the general trend is an increasing incidence with advancing age. This pattern is shown in more detail in Fig. 1. The overall incidence in younger age groups was low e.g. 0.8 per 100,000 in those 25–29 years. The marked increase in incidence was seen from 60 years upwards (200.86 per 100,000). A peak was reached at 70–74 years where the rate was 349.06 per 100,000 population and following a fall in the 75–79 year group the rate increased once more before declining in the 85+ years age group. While women reached their peak incidence at 70–74 years (210.53 per 100,000), men did not reach theirs until 80–84 years (694.12 per 100,000). The male to female incidence ratio rose steadily in each decade of age from a minimum of 0 in those aged 20–29 years through each decade—0.5, 1.9, 2.3, 2.5, and 2.9, respectively—to a maximum of 4.1 in those aged 80+ years (Table 2).

Table 3 shows comparative incidence data (all ages and by sex) for the four major regions of the British Isles: Northern Ireland, England and Wales, Scotland and the Republic of Ireland. The data presented for the UK are the latest years available (at the time of writing) and 1994 is the first year for which incidence data were available on a national basis in the Republic of Ireland.

In the years studied, Northern Ireland had the lowest incidence of lung cancer within the UK. The highest rates were observed in Scotland where, remarkably, the crude rate was 1.7 times that observed in Northern Ireland, 1.4 times that observed in England and Wales and 2.4 times that seen in the Republic of Ireland. Among women the disparities were marginally greater: the crude incidence rate among Scottish women at 70.3 (per 100,000) was 2.7 times that of women in the Republic of Ireland. Allowing for age did not change the regional ranking but had the effect of reducing the differences between the regions, especially between Northern Ireland and England and Wales where the age-adjusted person rates were 38.3 and 39.0, respectively. The low age-standardised rates obtained in Table 3 result from the structure of the (younger) World reference population chosen to facilitate International comparisons. Overall, both comparisons indicated that the rates in the Republic of Ireland were the lowest in the British Isles. However, it should be noted that the pattern of incidence in the Republic differed fundamentally from that observed in the other regions as the incidence rate of 40.8 (per 100,000) was lower than the corresponding mortality rate of 43.3. Such a finding may be due, at least in part, to under-ascertainment.

In order to compare the incidence of lung cancer in the four regions more formally, we adopted the model fitting approach described above. Using a linear logistic model we analysed the incidence in relation to three factors: Age, Sex and Region. A number of different linear logistic models were fitted to the regional data.

The null hypothesis of a common relationship between Incidence, Age and Sex in all four regions was tested. The regional homogeneity hypothesis was found to be untenable (\(\chi ^2\) = 850.4, df = 50, \(p<<\) 0.01). This result suggests that the pattern of incidence by age and sex is rather complicated and that it may be more appropriate to fit models separately for males and females and for each region. Accordingly, the simple model:in which the logarithm of the odds on incidence is a quadratic function of age was used.

Table 4 summarises the results obtained. The quadratic model provided an adequate description of the incidence observed amongst: (a) Northern Irish males and females, (b) Scottish females and (c) Southern Irish males. Formally, it narrowly failed to describe the trend in age incidence in Southern Irish females (\(\chi ^2\) = 10.4, df = 4, \(p < 0.05\), but almost all of the contribution to \(\chi ^2\) arose in one cell (20–29 yrs) in which 2 cases were observed, but only 0.27 were expected on the basis of the model. Ignoring this cell, the model also provides a satisfactory fit to the female data in the Republic.

The regression coefficients in Table 4 are all well identified statistically and their quantitative similarity is worth noting, particularly in relation to the lung cancer incidence data on the island of Ireland as a whole. Despite this, the model based on Eq. (2) did not explain the pattern of age-sex specific incidence observed in England and Wales nor among Scottish males. The observed and expected numbers of incident cases of lung cancer are shown in Table 5 for all 4 regions.

It will be noted that among males in England and Wales and in Scotland the model consistently and significantly: (a) over-estimated the incidence among those aged 50–59 years, (b) under-estimated the incidence in the next decade and (c) over-estimated the incidence among men aged 70–79 yrs. This pattern of lack of fit is also evident to a lesser (non-significant) extent in the incidence data for Northern Ireland, but is completely absent from the data observed in the Republic of Ireland. For females in England and Wales the model again failed to fit the observed numbers of incident cases aged 50–69 years (Table 5). However, the pattern of lack of fit differed somewhat from the male data observed in E and W.

From the outset methodological considerations were paramount and a comprehensive multi-source notification design was adopted in order to maximise the ascertainment of cases of lung cancer. The procedures implemented in this study were even more comprehensive—especially with respect to the number of sources polled and the amount of pro-active searching— than those currently being followed by the recently established NI Cancer Registry. Accordingly, very few cases, if any, were thought to have been missed.

The findings, that no single source was sufficiently complete and that only 137 of the 900 incident cases were known to all 4 sources during the course of the study constitute a rather remarkable indictment of the reporting systems which existed in Northern Ireland during 1991/1992. Thus, Table 1 should prove a valuable aid to the workings of the new registry. However, it is unlikely that the method of notification of cancer of the lung is peculiarly unique and the need for the NI Registry to pro-actively pursue a comprehensive (multi-source) notification strategy was clear.

The number of incident cases, at 900, exceeded the mortality by 112 cases. As noted earlier mortality had been used consistently by the DHSS and other health planners as a surrogate measure for incidence in the Province. The results show that this particular strategy under-estimated the true extent of the disease in the community by 12.45%. In the absence of temporal incidence data it was impossible to estimate the duration of such under-estimation, but the implications for funding and resource allocation in the Province would not be lost on specialists in the field.

Although, in the years studied, Northern Ireland had the lowest incidence of lung cancer in UK the overall rate was still 40% higher than the rate in the Republic of Ireland which, on these data, had the lowest rate in the British Isles. The absence of national incidence data in the Republic for 1991–1992 hampers the interpretation of the comparison. As we remarked earlier, the data for 1994 were the first to become available on a national basis in the Republic and some allowance must be made for potential under-ascertainment in the early years of the development of a new registry. Even so, it is highly unlikely that any such allowance would alter the Republic’s ranking in the regional comparison.

The null hypothesis of regional homogeneity in the age-sex incidence distribution was rejected. Despite this, we have demonstrated that a simple sex specific model in which the logarithm of the odds on incidence is a quadratic function of age fitted the Northern Ireland and Republic of Ireland data adequately. This model also explained the age-specific incidence distribution among Scottish females. However, it did not fit the data for England and Wales nor for Scottish males: i.e., the log-odds on incidence does not follow a simple quadratic form in age. It may be, of course, that a more general model would apply in each region and we recognise that factors such as smoking status, exposure to passive smoke, and socio-economic status have not been taken into account in this paper.

On a world-wide scale, the 1991–1992 Northern Ireland incidence rate of 57.0 per 100, 000 population was intermediate, being lower than some countries—England and Wales, Scotland, USA, and Canada—but, higher than others—Australia, India, Israel and Thailand [5‐8]. Neither was the male to female incidence ratio of 2.1 to 1 particularly unusual. Data from the International Agency for Research on Cancer [3] showed marked inter-country variation in male to female incidence ratios, ranging from less than 3:1 (USA, Denmark and Israel) to those in excess of 9:1 (Germany, France and Italy). Accordingly, it seems that individuals in Northern Ireland were not at particularly high risk of developing the disease in 1991/92.

In 2021, the epidemiological picture is rather different and the purpose of this new section is simply to bring the regional incidence results up to date.

Examining current Registry data [29‐33], we were able to compile Fig. 2 which shows the regional trends in crude incidence rates for the period 1993–2018. Not all Registries have data available for the whole period: for example, the Welsh Registry has only published data from 2002.

The pattern which emerges is hardly re-assuring, all of the trends are increasing. The figure does, however, largely confirm the 1991–1992 ranking (highest–lowest) of Scotland, England, Northern Ireland and the Republic and confirms the conclusion that the position of ROI would remain unchanged, when its registry was fully operational.

It may be argued that the increasing trends are due to aging populations. Whilst this may be true in part, it seems an unlikely explanation for the trend in NI, where there was a shallow decline after 1991–1992 until c2000. Thereafter, a gradual increasing trend was experienced until 2010, when the increase accelerated up to the higher rates observed in 2016–2018. The 2017 rate in NI is c72, an increase of 26.3% over the 1991–1992 rate (57.0, Table 3) and it has been driven by a dramatic rise in incidence among females (not shown). The most striking finding is that the incidence rate in NI exceeded the rate in England in 2016! In 1991–1992 epidemiologists would not have thought such a result possible.

These increases in NI and elsewhere are the more puzzling, when considered against a backcloth of successful smoking bans in public places and declining population smoking prevalence rates in the UK, down from 20.2% in 2011 to 14.1% in 2019, [34], overall, and in the Republic, down from 28.3% in 2008 (last high) to 15.0% in 2020 (HSE, 2020) [35].

Thus, it seems that the conclusion reached about NI’s position in 1991–1992 was rather optimistic and further studies are now required to identify the factors underpinning these increasing trends.