Introduction
Previous analysis of data from cancer networks throughout the UK has shown the highest incidence of laryngeal cancer to be in parts of Scotland and Northern England, with lowest rates being recorded in Southern England. 1 The population served by the Humber and Yorkshire Coast Cancer Network has one of the highest incidence rates of laryngeal cancer in the country amongst older males (aged 50–74 years). 1 , 2 A 5-year epidemiological study performed by the local health authority identified up to a 10-fold difference in laryngeal cancer incidence rates between 106 electoral wards within the Humber and Yorkshire Coast Cancer Network.Reference Sethi, Rafferty, Rawnsley and Jose 3 The electoral wards with the lowest incidence reported rates of 3 per 100 000; in contrast, incidence rates of 25–30 per 100 000 were recorded for the wards of Myton and Marfleet in Hull. 2 , Reference Sethi, Rafferty, Rawnsley and Jose 3 Public health data indicated that Myton ranked in the top five electoral wards for risk factors associated with laryngeal cancer, which included binge drinking levels, smoking prevalence and social deprivation.Reference Sethi, Rafferty, Rawnsley and Jose 3
A public health campaign targeting specific high-risk electoral wards in the Hull and Grimsby areas was launched in September 2011 and lasted for three months. The campaign employed a multi-modal public engagement approach, which included radio, television, internet and poster media coverage. Further details of the campaign can be found online. 4 It was thought that the public health campaign would improve knowledge and health-seeking behaviour, and prompt earlier presentation and diagnosis of laryngeal cancer, thus improving survival outcomes. Existing publicised analysis on the effectiveness of the campaign (cancer awareness measure and recall) has shown a limited effect.Reference Sethi, Rafferty, Rawnsley and Jose 3 This study specifically assessed the early impact of the public health campaign on the stage of laryngeal cancer at presentation in the Humber and Yorkshire Coast Cancer Network after September 2011 to the present.
Materials and methods
Data collection
Details of patients presenting with laryngeal carcinoma within the Humber and Yorkshire Coast Cancer Network were obtained from the regional cancer database and the Data for Head and Neck Oncologists (‘DAHNO’) registry for January 2009 to July 2014 inclusive (67 months in total). This period included the dates of the public health campaign, which was conducted from September 2011 for three months.
Stage of laryngeal cancer at presentation was categorised as either early (tumour stage T1 and T2) or late (T3 and T4) disease. Laryngeal cancer staging followed the standardised consensus technique for head and neck squamous cell carcinoma. 5 Trends related to tumour stage at presentation were evaluated, particularly in relation to the public health campaign. The findings were statistically analysed using segmented regression analysis of interrupted time series data.
Statistical analysis
Segmented regression analysis of interrupted time series data allows assessment, in statistical terms, of how much an intervention changed an outcome of interest, immediately and over time.
The segmented regression model is shown below, where ‘Y t ’ is the outcome in month ‘ t ’, ‘β0’ estimates the baseline level of the outcome at time zero, ‘β1’ estimates the change in the outcome that occurs with each month before the intervention (i.e. the baseline trend), ‘β2’ estimates the level change in the outcome immediately after the intervention, that is, from the end of the preceding segment, and ‘β3’ estimates the change in the trend in outcome after the intervention, compared with the monthly trend before the intervention. The error term ‘et’ at time ‘t’ represents the random variability not explained by the model.
$${\rm Y}_t = {\rm \beta} _0 + {\rm \beta} _1 \times {\rm time}_t + {\rm \beta} _2 \times {\rm intervention} + {\rm \beta} _3 \times {\hbox{time after intervention}}_t + {\rm et}$$
To assess the fit of the final model, we examined residuals around the predicted regression lines. The partial auto-correlation function and auto-correlation function residual plots were examined. The Durbin–Watson statistic was used to assess remaining auto-correlation. The Durbin–Watson statistic, reported by most least squares regression programs, tests for serial auto-correlation of the error terms in the regression model. Values close to 2 indicate no serious auto-correlation. Statistical significance was set at p < 0.05.
Ethical considerations
The public health campaign was registered with the Hull and East Yorkshire Research and Development Committee.
Results
A total of 198 cases of laryngeal cancer were diagnosed in the Humber and Yorkshire Coast Cancer Network during the study period (January 2009 to July 2014). Prior to the public health campaign, 114 cases of laryngeal cancer were diagnosed, whereas 84 cases of laryngeal cancer were diagnosed following (and inclusive of) the campaign. The number of early tumour stage (T1–2) laryngeal cancers diagnosed totalled 46 pre-campaign and 46 post-campaign. The number of late tumour stage (T3–4) laryngeal cancers diagnosed totalled 68 pre-campaign and 38 post-campaign.
Early stage cancer
Regression analysis findings for early stage (T1–2) laryngeal cancer patients are shown in Table I and Figure 1. Prior to the intervention, there was an annual rise of 0.039 in the number of cases, but this was not statistically significant (p = 0.115). There was no statistically significant change in the number of cases immediately after the intervention (p = 0.191).
Fig. 1 Regression analysis for early tumour stage (T1–2) patients.
Table I Segmented regression analysis findings for early and late stage laryngeal cancer patients*
* In relation to the public health campaign. SE = standard error
Following the intervention, there was a rise of 0.013 (0.039 − 0.026) in the number of cases per month. These trends were not statistically significant (p = 0.443).
Late stage cancer
Regression analysis findings for late stage (T3–4) laryngeal cancer patients are shown in Table I and Figure 2. Prior to the intervention, there was an annual fall of −0.036 in the number of cases, but this was not statistically significant (p = 0.052). There was no statistically significant change in the number of cases immediately after the intervention (p = 0.680).
Fig. 2 Regression analysis for late tumour stage (T3–4) patients.
Following the intervention, there was a decrease of −0.028 (−0.036 + 0.008) in the number of cases per month. These trends were not statistically significant (p = 0.767).
Discussion
Main findings summary
The results show no statistically significant differences in the presentation of early (T1–2) and late (T3–4) laryngeal cancer cases following the public health campaign. Prior to the campaign, the annual number of early stage (T1–2) laryngeal cancer presentations were steadily increasing (+0.039), whereas the number of late stage (T3–4) presentations were falling (−0.036). These numbers were not statistically affected by the campaign.
Monthly rates of early (T1–2) and late (T3–4) stage laryngeal cancer increased (+0.013) and decreased (−0.028) respectively in relation to the public health campaign; however, the rates were not statistically correlated with the intervention.
Although our study did not identify any significant correlation, it is interesting and useful to highlight the current trends in laryngeal cancer for the region. Our analysis may serve as a useful barometer into the ongoing trend in stage of laryngeal cancer at presentation (on a larger scale) and may help reflect the current status of public health awareness. These trends may be associated with evolving awareness from growing local, regional and national media coverage, access to social media, and existing public health campaigns for detecting cancer.
Some of the findings observed may be expected following a public awareness campaign, particularly in the early period, where participants would be expected to seek medical attention for key symptoms highlighted in the media. Theoretically, rates of early cancer detection would be expected to rise, as was the case with the findings observed in our study. However, given the pre-existing trend in cancer detection rates (for early and late cancers, as shown in Figures 1 and 2), it is difficult to comment on the role of such a short-lived campaign at this current stage. It is more likely that such a limited public health campaign would lead to changes in the presentation of patients to general practitioners and in urgent two-week wait referrals to specialists. Previously published research on this campaign, by Sethi et al., revealed an increase in the numbers of individuals presenting to the general practitioner with throat cancer symptoms (during the peak of campaign activity and after the campaign).Reference Sethi, Rafferty, Rawnsley and Jose 3 However, the findings were not significant when compared to a control group (an appropriately matched electoral ward). Despite an apparent increase in two-week wait referrals, the authors concluded that, given the pre-existing trend of year-on-year increase in referrals, it would be difficult to fully attribute the changes observed to the public health campaign.Reference Sethi, Rafferty, Rawnsley and Jose 3
Implications
Given the complexity of public health campaigns, investigation of their effectiveness is often difficult. In order to fully evaluate the impact of the campaign, the ‘credibility, completeness and transferability’ of the evidence should be examined.Reference Rychetnik, Frommer, Hawe and Sheill 6
Although it is not within the scope of this article to discuss the effectiveness and limitations of this public health campaign, it is evident that there are several factors which directly or indirectly effect interpretation of the data derived from it. The lack of statistical significance of our study raises several questions. For instance, has the research failed to identify an effect when one truly exists (evaluation failure)? Or is there truly no effect (programme failure)?Reference Hawe 7 Increasing the follow-up period may help further evaluate latent (post-campaign) responses.
Programme failure may be considered an inadequacy of the campaign itself or of its implementation. The public health campaign was targeted towards high-risk electoral wards in the Humber and Yorkshire Coast Cancer Network identified from preliminary epidemiological data. Our study looked retrospectively at the total number of laryngeal cancer patients presenting to the Humber and Yorkshire Coast Cancer Network, and not the number of patients by electoral ward. Therefore, a true reflection of campaign effectiveness may not have been identified from our analysis, as the results may have become ‘diluted’ by the regions not exposed to the intervention. A more accurate picture may be obtained if laryngeal cancer stage at presentation by electoral ward was assessed, after the intervention. However, one could counter-argue that a significant change in cancer diagnosis might be expected, as a larger number of laryngeal cancers would arise from those geographical areas at highest risk, thus increasing overall cancer detection numbers. In addition, the effect of the campaign may not truly be localised to the electoral ward, given the wider-reaching effects of media coverage (e.g. radio, sports events and internet advertisement) and local migration. These issues make assessment of public health campaign effectiveness difficult.
Sethi et al. suggested that inadequate implementation could relate to several factors, including campaign length and limited use of the most effective medium (television).Reference Sethi, Rafferty, Rawnsley and Jose 3 It is more than likely that a single three-month period represented too short a timeframe to instigate a significant change in behaviour and ultimately stage of laryngeal cancer at presentation for the region. This theory may be supported by previously published findings from a post-campaign survey, which showed limited improvement in awareness and a rapid decline in recall.Reference Sethi, Rafferty, Rawnsley and Jose 3
-
• This paper describes the first public health campaign for laryngeal cancer performed in the UK
-
• Findings suggest limited changes in stage of laryngeal cancer at presentation in relation to the campaign
-
• Public health campaigns are complex processes
The relatively small numbers in our study have implications for the findings presented. They may have affected the ability of analysis to detect significance in relation to the public health campaign, especially if individual wards are considered. Future analysis of a larger sample size will enable greater statistical power, and there is also the possibility of combining the results of similar studies.
Our findings suggest that the ‘croaky voice’ public health campaign had a limited effect upon the number of presentations of laryngeal cancer in the Humber and Yorkshire Coast Cancer Network. It is difficult to currently attribute any of the changes observed in our analysis to the campaign given the complexities associated with the design and assessment of public health interventions. For this reason, the several factors discussed must be taken into account in any future public health campaign targeting populations with head and neck cancer.
Conclusion
The ‘croaky voice campaign’ was the first of its kind to address laryngeal cancer in the UK. Our evaluation of the initial results has shown trends to suggest limited alterations in the rates of early (T1–2) and late (T3–4) stage laryngeal cancers relative to the campaign. However, at this stage, the relationship remains unclear. Several key factors have been described regarding performing public health campaigns and evaluating the results, which must be considered when planning future studies. These include the use of more effective media channels, longer intervention implementation and a prolonged follow-up period.
