Invasive meningococcal disease in England: assessing disease burden through linkage of multiple national data sources

Invasive meningococcal disease (IMD) is associated with significant case fatality and long-term morbidity despite availability of effective antibiotics and intensive care support in industrialised countries [1, 2]. Efforts to control the disease have, therefore, focussed on prevention, primarily through vaccination [3]. Most cases of IMD are caused by one of six meningococcal capsular groups, which are characterised by their unique polysaccharide capsule (A, B, C, W, X or Y) [4]. In the United Kingdom, meningococcal group C (MenC) is rare because of a successful national immunisation programme and group B (MenB) is responsible for most IMD cases across all age groups [5].

In England, national surveillance of laboratory-confirmed IMD is conducted by Public Health England (PHE), through the Meningococcal Reference Unit (MRU), which provides a national service for species confirmation, grouping, typing, subtyping and antimicrobial susceptibility testing of all invasive Neisseria meningitidis isolates. The MRU also provides free non-culture PCR confirmation of meningococcal diagnosis (including genogroup and genosubtype analysis) for clinical specimens that are routinely submitted by National Health Service (NHS) hospitals in England [5]. We have recently highlighted the added value of the PCR-testing service because of its high sensitivity and specificity, which has proved invaluable in achieving high case ascertainment rates for IMD surveillance [6]. However, whilst the current surveillance provides important information on the epidemiology of IMD and trends over time, it may underestimate the true burden of disease if clinical specimens are not submitted to the MRU or if hospitals submit their clinical isolates to private laboratories that provide a similar service. Cases that are diagnosed clinically without laboratory-confirmation, too, would not be included in the current surveillance. This may be more important for the less prevalent meningococcal capsular groups, which have more varied clinical presentations and often affect older adults with co-morbidities [7]. Likewise, meningitis cases may be missed if lumbar punctures are not performed [8].

In order to define the true burden of IMD, it is important to evaluate routine surveillance against alternate data sources. Accurate data form an essential component for modelling the potential impact (including cost-effectiveness) of any new vaccine which, in turn, can be used to inform national vaccination policy. The use of existing national datasets as a means of undertaking health research and improving public health monitoring offers exciting opportunities to undertake very large scale studies at very low cost. This study aimed to integrate epidemiological, clinical, microbiological and outcome data using multiple national datasets to assess the burden of IMD in England over a five-year period.

PHE conducts enhanced surveillance of vaccine-preventable infections in England. For this study, five independent national datasets covering the five-year period during 2007–2011 were de-duplicated prior to linkage (Table 1). In addition to data collected through provision of a national reference laboratory service, PHE also routinely receives electronic notifications of clinically significant isolates by NHS laboratories through LabBase2 and reports of PCR-confirmed cases from private laboratories that support microbiology services for NHS hospitals. PHE also has access to Hospital Episode Statistics (HES), an online database containing electronic records of patients admitted to NHS hospitals, and receives annual electronic reports of all death registrations in England from the Office for National Statistics (ONS). PHE has been granted access to these data, through Regulation 3 of The Health Service (Control of Patient Information) Regulations 2002 (http://​www.​legislation.​gov.​uk/​uksi/​2002/​1438/​regulation/​3/​made) in order to fulfil its legal responsibility to monitor the impact, safety and effectiveness of the national immunisation programmes.

Prior to linkage, the Hospital Episode Statistics (HES) spells and episodes were converted to individual admissions. The electronic Patient Demographic Service (PDS) was used to maximise the number of cases with an NHS number using other identifiable data for individual cases. PDS is a national electronic database of NHS patient demographic details that allows healthcare staff to identify patients quickly and accurately (http://​systems.​hscic.​gov.​uk/​demographics/​pds). For all interrogated datasets, cases linked by NHS number were verified using the other linkage parameters (date of birth, sex, etc.) before being considered exact matches. The remaining unmatched cases where then matched using date of birth, gender and postcode, followed by different combinations of approximate date of birth, gender, region and approximate time of infection. All cases were then linked with the ONS death registration records to identify any extra cases not reported in the MRU or HES dataset, to ensure more complete outcomes for linked cases and to ascertain cause of death. Because LabBase2 and the private laboratory provided very few additional laboratory-confirmed cases to the total MRU-confirmed cases, these two data sources were interrogated at the end to identify any HES cases that might have been confirmed by laboratories other than the MRU. For LabBase2, individual clinical specimen records with an organism code for NEISSERIA MENINGITIDIS and with a specimen date in the study period were extracted, de-duplicated and linked to the MRU database. Linkage was done using NHS Number, patient surname, forename, date of birth, date of specimen, postcode and reporting laboratory. Any remaining episodes that did not link to the MRU database were then linked to the HES database. Case fatality ratio (CFR) refers to a fatal outcome within 30 days of a positive laboratory test and, where this date was not available, then the date of admission was used as a proxy. Patients with the same postcode for place of residence and place of death were considered to have died at home.

The linkage of multiple national data sources has provided useful insight into the total burden of IMD in England. The provision of free, national PCR-testing of clinical samples in addition to species confirmation and capsular grouping of clinical isolates by the MRU has not only guided clinical management of patients, but also improved case ascertainment and completeness of IMD surveillance [6]. LabBase2 and private laboratory reports added only a few more cases to the total number of confirmed IMD reports by the MRU, thus confirming two recent publications which reached the same conclusions [5, 6].

We successfully linked 4275 MRU-confirmed cases over the five-year period to a hospitalisation record in HES, where IMD (A39*) or bacterial meningo-encephalitis (G00*) was recorded. The remaining 840 (16 %) MRU-confirmed cases did not have sufficient identifiers for successful linkage, had non-specific infection-related ICD10 codes in HES, or died prior to hospital admission – either in the Emergency Department or outside hospital (e.g. at home).

At the same time, interrogating HES for IMD admissions identified an additional 2792 cases but more than half matched to an MRU PCR-negative test-result. Cross-checking these HES+/MRU- cases with LabBase2 and private laboratory reports identified only ten additional cases with positive laboratory confirmation. Discrepancies between the positive LabBase2/private laboratory testing and negative MRU PCR-tests may be due to the timing of sample submission or in the sensitivity of the diagnostic tests used. Overall, however, almost 1500 patients (equivalent to 300/year) had an IMD diagnosis in HES with a PCR-negative test by the MRU.

Of the remaining 1327 HES cases that did not match any MRU-tested samples, only 73 had additional laboratory-confirmation through LabBase2 or the private laboratory over the five-year period. This group also had the lowest proportion of identifiable information for linkage. It is possible that, if sufficient identifiers had been available, at least some of MRU-positive cases that did not link to HES might have linked to these HESonly cases. This would, however, still not account for >1000 HESonly cases (equivalent to >200/year) without any laboratory-confirmation for IMD.

The characteristics of the HESonly cases – age distribution, seasonality, short duration of in-patient stay (≤48 h in 40 %) and CFR – resembled HES+/MRU- cases more than MRU+/HESA39+ cases. In clinical practice, children (and many adults) often present with non-specific symptoms and signs such as fever and petechial rash and these cases usually end up being coded in HES as IMD because of lack of an alternative, more conclusive diagnosis. At the same time, however, some of these patients will have clinically-diagnosed IMD cases without laboratory-confirmation (e.g. those who received antibiotics prior to blood culturing and did not have a sample submitted for PCR-testing), although it is difficult to estimate the contribution of such cases to total HESonly cases.

In 1998, when the UK was experiencing a national MenC outbreak, we had estimated that there were 46 % more A39+ cases recorded in HES (n = 3316) than confirmed by the MRU (n = 2,276) [9]. This was a crude estimate of under-ascertainment because individual-level data-linkage was not performed. The current study has demonstrated the additional value of linking multiple national data sources whilst highlighting the complexities of linking large datasets that were not designed for integration. A straightforward comparison would indicate 32 % more HESA39+ (n = 6727) than MRU-confirmed (n = 5115) cases during 2007–11, which is lower than 46 % reported in 1998, possibly because of a higher proportion of cases with laboratory testing and confirmation [6]. Linkage of individual patient records, however, showed that a significant proportion of MRU-confirmed cases were not coded as IMD (A39+) in HES or died before admission to hospital. At the same time, a substantial proportion of hospitalised patients with a HES A39+ discharge code were either identified as PCR-negative by the MRU or were not laboratory-confirmed. During 2007–11, therefore, the latter group comprising 1254 HESonly cases would add 25 % more cases to the 5115 MRU-confirmed cases compared to our previous estimate of 46 % excess cases in 1998 [9].

Increasing use of NHS numbers for individual patients was critical in facilitating successful linkage, as did access to the patient demographic service (PDS) batch-testing which provided NHS numbers for cases with partial identifiers, highlighting the need for a unique identifier across all data sources for future linkage. For cases with missing NHS number, linkage success fell rapidly, making it difficult to ascertain whether the additional unlinked records are genuinely more cases or simply unlinked duplicates. Despite the wealth of data generated, it is clear that linkage of national datasets in their current format is labour-intensive and not amenable to automation, but could be repeated at regular intervals to monitor trends over time.

Our analysis confirms that the MRU captures nearly all laboratory-confirmed IMD cases and is, therefore, ideal for monitoring national trends in near-real time. Linkage with HES, however, identified an additional 2792 hospitalised cases over 5 years with an IMD diagnostic code, including half that linked to an MRU PCR-negative result; of the remaining cases, it is difficult to estimate what proportion were genuine but unconfirmed IMD cases. Reassuringly, though, these additional HES cases indicate increased awareness of IMD among clinicians with low thresholds for hospitalising patients with suspected IMD. Following the introduction of a broad-spectrum meningococcal vaccine (4CMenB; Bexsero, GSK Biologicals) into the UK childhood immunisation programme in September 2015, monitoring trends in laboratory-confirmed and unconfirmed hospitalised IMD cases could provide useful insight into the true burden of IMD. In the meantime, our results justify the continued use of MRU data for monitoring IMD epidemiology in England.