Measles and Pertussis outbreaks in England and Wales: a time-series analysis

Introduction

Vaccinations are one of the most effective public health interventions¹. The UK’s childhood vaccination programme is offered up to the age of 5, protecting against 13 infectious diseases². Since 2014, there has been a decreasing trend in vaccine uptake, which has been exacerbated by the COVID-19 pandemic³. In addition to vaccine uptake, the timeliness of the administration is also critical to prevent outbreaks, particularly for Measles and Pertussis, where evidence suggest delays in administration of the first dose are the biggest predictors of non-completion of the vaccination schedules⁴. The number of children at 24 months who have completed their first dose of the Measles Mumps and Rubella (MMR) vaccine has recently fallen to 89.2%, down from 90.3% the previous year (with coverage falling below 90% in 61 of the 149 Local Authorities) and those completing the second dose by age 5 down nearly 1% from the previous year⁵. Between 2022–2023 the prenatal vaccination rate for Pertussis was 60.7%, representing a 4% drop from the year ending 2022 and 7.1% decrease from 2021⁶. Vaccine rates for Measles and Pertussis in the UK are below the 95% target set by the World Health Organisation⁷.

As a consequence of decreasing vaccination rates, waning vaccine protection and the high transmissibility of both Measles and Pertussis, cases have been steadily rising in England and Wales⁸. In the last 6 months of 2023 there were 1459 cases of Pertussis and 999 cases of Measles, compared to 428 and 658 in the first 6 months, respectively. These data are also substantially higher compared to the same period in 2022 (337 and 397) and in 2021 (296 and 249)⁹.

Early identification of outbreaks is critical. Firstly, to implement targeted interventions, preventing further transmission, protecting vulnerable populations, and reducing mortality and morbidity (including reducing secondary infection risk)¹⁰. Secondly, it provides a platform for communication and public awareness efforts. Rapid dissemination of information to healthcare providers, communities, and the public is critical for promoting preventative measures, encouraging vaccination, dispelling misinformation - contributing to the broader goal of achieving and maintaining population immunity¹¹. There is a critical need to explore additional mechanisms to traditional health surveillance techniques, particularly those operating in real time, to monitor cases or even predict outbreaks.

There is a growing body of research evidencing of the utility of online health information searching (OHIS) for health research across several domains, exploring; the impact of major disease awareness programmes^12–14 the relationship between search data and disease burden measures¹⁵ and how outbreak notifications impact OHIS¹⁶. OHIS has also been used to predict and monitor disease outbreaks in combination with traditional health surveillance systems and has been shown to accurately predict cases of COVID-19¹⁷, HIV¹⁸, Zika¹⁹, self-harm²⁰ and influenza²¹. However, OHIS has not been used to predict outbreaks of childhood infectious diseases, such as Measles and Pertussis in the UK.

In this exploratory time-series study, we explored relationships between OHIS for Measles and Pertussis and cases in the context of the current outbreaks in England and Wales, to investigate whether Google Trends (GT) offers a predictive utility as a health surveillance tool.

Methods

Results

Cases and Google Trends data

The total number of cases for Measles and Pertussis over the 52-week study period (01/01/2023 – 07/01/2024) and the average number of cases per week are presented in Table 1. Measles and Pertussis cases with the RSV for “Measles” and “Pertussis” are illustrated in Figure 1. The RSV for other search terms across the study period are presented in Figure 2.

Table 1. Measles and Pertussis cases for the study period.

	Number of cases	Mean cases per week (SD)	Range (min-max)
Measles	1657	31.87 (15.14)	(9-75)
Pertussis	1887	36.29 (37.63)	(6-168)

SD, Standard Deviation’ Min, Minimum; Max, Maximum.

Figure 1. Measles and Pertussis cases with realtive search volumes for the study period.

Figure 2. Relative Search Volume for all search terms across the study period.

Discussion

We explored the relationship between OHIS and case data for Measles and Pertussis to assess whether the use of key search terms could support public health surveillance or even predict outbreaks of both diseases. Our analyses strongly suggests that GT provides a predictive utility for the increases in cases for Measles and Pertussis in advance of diagnosis based on disease and symptom searching. For Measles cases, the search term “measles” significantly predicted subsequent cases 1 week later, searching for “spots” offered week predictive utility for cases 3 weeks later. For Pertussis cases, the search term “Pertussis” offered a less robust association with Pertussis cases, however this is likely attributable to it being a less known term in the community. However, the more informal terms “whooping cough” and “100 day cough” showed a much more significant predictive relationship, but the strongest association was found for “cough”. Increased searching for cough was significantly predictive of Pertussis cases 3 weeks later and remained the most significant predictor of cases in multivariable analysis.

There are limitations that need to be considered. Firstly, other events, such as news reports may have impacted searching trends, particularly in the context of Measles and Pertussis outbreaks, however time series analysis in this study suggests that OHIS was occurring before the diagnostic and disease notification process. Secondly, this study only used Google searching, missing out on searching using other platforms, however Google remains the most frequently used search engine in the UK (93.69% of devices)³⁰. Thirdly, calculations applied to obtain RSVs are not publicly available and applied algorithms may obscure the observed trends, however, previous evidence has shown the data to accurately predict other infectious disease outbreaks¹⁸ and be comparable to the United States Centre for Disease Control in predicting the timing of influenza outbreaks in the US²¹. Currently, the provision of location specific searching provided by GT is inadequate for the UK, so data pertain to the national level only. Being able to divide search data geographically, perhaps by county or Integrated Care Board would enhance the predictive utility of GT and elucidate the locations of cluster outbreaks that require urgent response. This would allow for the inclusion of vernacular and terminology specific to different geographies. Finally, this study is retrospective, protocols using GT for health surveillance techniques and outbreak monitoring in real time need to be developed and tested.

This is the first study to explore the predictive utility of GT for measles and Pertussis in England and Wales, however findings are consistent with previous evidence supporting the health surveillance utility of OHIS data. Samaras et al. employed a gamma distribution model showing how OHIS predicts outbreaks of Scarlet Fever 5 weeks before the peak using symptoms and scarlet fever associated search terms³¹. Furthermore, GT data was utilised by Prasanth et al alongside modelling techniques to accurately forecast COVID-19 incidence, cumulative cases, and deaths at the national level during the pandemic³². GT data has also been used to accurately predict Norovirus cases in the UK and more accurately than other real time data searching sources, including Wikipedia³³. Internationally, GT data has been used to predict infectious disease outbreaks at more local levels, forecasting malaria, dengue fever, chikungunya, and enteric fever outbreaks in two districts in India³⁴. In the US, GT has been shown to accurately predict HIV case numbers¹⁸ and mpox outbreaks by state³⁵ as well as influenza cases and hospitalisations³⁶.

In addition to changes in trends for disease cases reflecting earlier changes observed in OHIS, the converse was also shown in our time series analysis, where OHIS for Measles and Pertussis were subsequently predicted by increases in cases. This is likely attributable firstly to increasing cases over time and secondly, as those receiving diagnoses begin to search for disease-related or symptom management information³⁷. This is an important finding illustrating the need for the provision of high-quality, evidence-based public health related information to support communities with transmission prevention techniques, healthcare seeking guidance and to prevent the spread and impact of misinformation¹⁶. Additionally, it is expected that subsequent engagement with Measles and Pertussis topics take places via alternative digital platforms. Mahroum et al prosed a stimulus-awareness-activism framework, where surges in OHIS are followed by further online and offline discourse and has been evidenced by increased Twitter/X discussion on Chikengunya following outbreaks³⁸.

Not all symptom searching offered the same predictive utility for example “fever” searches offered little predictive utility for Measles cases. This is an important finding as this may reflect how different symptoms present at different times across the course of the disease and possibly how more troubling symptoms lead to higher levels of searching³⁹. Future work exploring GT as a health surveillance tool must identify the symptoms for which searching most accurately associates with case data.

Outside of increasing vaccine uptake in the community for Measles and Pertussis, additional real time surveillance tools to increase the likelihood of the early identification of case increases for Measles and Pertussis is of critical importance. This study suggests OHIS data, as provided by GT, can work as an alternative or adjunct to more traditional health surveillance systems (Royal College of General Practice Weekly Returns, NOIDs etc). Early detection allows for targeted interventions, preventing further transmission, protecting vulnerable populations in the community, and reducing mortality and morbidity¹⁰.

Conclusion

Our findings suggest that trends in online health information seeking for Measles and Pertussis precede trends in cases in England and Wales, indicating that GT may provide an additional, real time data source with predictive utility in detecting outbreaks of Measles and Pertussis. Further work is required to explore how GT data can be used in conjunction with other, traditional health surveillance systems to monitor or even predict disease outbreaks, to better target public health interventions.

Ethics and consent statement

No Ethics and consent required for the performed study