INTRODUCTION
Meningococcal disease, caused by infection with the bacterium Neisseria meningitidis [Reference Granerod1], is a life-threatening disease which may be associated with a rapidly progressive haemorrhagic rash, meningitis, septicaemia and death. Rarer forms of the disease include conjunctivitis, otitis media, epiglottitis, urethritis, arthritis and pericarditis [Reference Rosenstein2]. Children aged <5 years have the highest rates of notification for meningococcal disease [Reference Ladhani3] and the highest death rates (0·58 deaths/100 000 Australian population in 2005–2007 [Reference Chiu4]).
In Australia in 2011, nearly 84% of meningococcal disease cases were serogroup B, with serogroup C accounting for only 4% [Reference Lahra and Enriquez5]. Since the introduction of a vaccine for group C meningococcal disease in 2003 [Reference Cohen6] there has been a significant and sustained reduction in cases of group C disease [Reference Lahra and Enriquez5]. Greater prevention of meningococcal disease will not be achievable without a vaccine against group B disease [Reference Girard7]. An expensive serogroup B vaccine has been approved for use in Australia [8], but it is not yet approved for funding under the National Immunization Programme. This same vaccine has been rejected for routine use in the UK because of low and uncertain cost-effectiveness [Reference Moxon and Snape9], while restricted use in children at a cost-effective price is being negotiated [Reference Pollard, Riordan and Ramsay10].
Accurate reporting of meningococcal disease is essential in order to monitor the impact and cost-efficacy of public health policy including vaccination programmes, and provide an appropriate public health response to cases and their close contacts [Reference Breen11, Reference de Greeff12]. Yet, meningococcal disease poses some unique challenges for monitoring systems. It is a rare disease; in 2012 only 65 cases were reported in New South Wales (NSW), Australia (population over 7 million) [13], therefore many person-years of data are needed to accurately estimate its incidence. Moreover, infection can progress from initial symptoms to death within hours, and many early cases of meningococcal disease present with only non-specific symptoms [Reference Thompson14], which can lead to both misdiagnosis and over-diagnosis [Reference Nadel15]. Many cases are referred to specialist centres for assessment, which can complicate hospital meningococcal diagnosis reporting. These challenges mean that even the most rigorous of recording systems, such as notifiable disease registers, can be inaccurate [Reference Stephen16]. Paediatric meningococcal diagnoses might be particularly prone to higher rates of false-positive reporting [Reference Breen11]. Accurate estimates of disease incidence must both minimize case under-ascertainment and quantify the proportion of false-positive diagnoses [Reference Orton, Rickard and Miller17].
In NSW, several databases record cases of meningococcal disease. These include notification data (the Notifiable Conditions Information Management System), hospitalization data (the Admitted Patient Data Collection), and mortality data (from the Australian Bureau of Statistics; ABS). The current study used linked notification, hospitalization and mortality data from 2000–2007 inclusive for all NSW children aged 0–14 years to: (i) estimate rates of false-positive meningococcal disease diagnoses in hospital records; (ii) quantify case under-ascertainment using capture–recapture estimation methods; and (iii) estimate the incidence of meningococcal disease, adjusted for false-positive reporting and case under-ascertainment.
METHODS
Databases and data linkage
This study used four health and administrative datasets:
-
• The NSW Perinatal Data Collection holds records of all births in NSW, and was used to calculate the population at risk for incidence rate calculations. Records for live births from 1 January 1994 to 31 December 2007 were used.
-
• The NSW Notifiable Conditions Information Management System is a register of cases of infectious and other conditions specified in the NSW Public Health Act 2010, including confirmed and probable cases of meningococcal disease, that are notified to the NSW Ministry of Health. Confirmed cases of meningococcal disease require either (1) laboratory-confirmed evidence or (2) laboratory suggestive evidence in combination with clinical evidence. Probable cases require only clinical evidence, including the absence of evidence for other causes of clinical symptoms and either (a) clinically compatible disease or (b) clinically compatible disease and recent close contact with a confirmed case [Reference Chiu4]. Cases are recorded as a particular serogroup, serogroup ‘NOS’ (‘not otherwise specified’) or serogroup ‘not typable’. Records for cases for NSW-resident children (0–14 years) with estimated disease onset dates from 1 July 2000 to 31 December 2007 were used.
-
• The NSW Admitted Patient Data Collection records all inpatient separations (discharge, change of service category, death or transfer) from all public, private, psychiatric and repatriation hospitals in NSW. Meningococcal disease was identified on the basis of the International Statistical Classification of Diseases and Related Health Problems, Tenth revision (ICD-10) diagnostic codes A39·0–A39·9 inclusive in any diagnosis field. Diagnoses are coded for each new separation. A patient may have a number of separations corresponding to a single episode of illness if they are transferred between hospitals, have a change of service category within a single hospital, or are admitted several times. Separations were considered to be part of the same illness episode if no more than 24 h elapsed between discharge from one hospital separation and admission into the next. Admission dates of 1 July 2000 to 31 December 2007 were used for NSW-resident children aged 0–14 years.
-
• The Australian Bureau of Statistics collates coronial information and mortality data from the State Registry of births, deaths and marriages. Deaths from meningococcal disease were identified on the basis of ICD-10 codes A39·0–A39·9 inclusive in any of the cause- of-death fields. Records for deaths of NSW-resident children (0–14 years) occurring between 1 July 2000 and 31 December 2007 were used.
The datasets were linked together probabilistically by the NSW Centre for Health Record Linkage (CHeReL) using personal identifiers, including full name, date of birth, sex and address. The CHeReL's probabilistic linkage procedures are designed to achieve around 5/1000 incorrect links and 5/1000 missed links [18]. Following data linkage, each child (with a unique identifier) had a set of data records which could include notification/s, hospitalization/s, birth and death records. Only de-identified data were analysed by researchers.
Ethical standards
Ethics approval was received from the NSW Population and Health Services Research Ethics Committee (Ref. No. 2009/11/193) and the University of Western Sydney Human Research Ethics Committee (Ref. No. H10651). The authors assert that all procedure contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Study population
The study population was all NSW-resident children aged between 0 and 14 years (a mean of 1 324 925 children/year) who had the potential to have notification, hospital separation or death recorded between 1 July 2000 and 31 December 2007. The Perinatal Data Collection recorded a total of 1 223 313 children born alive in NSW between 1 January 1994 and 31 December 2007.
Analyses
All analysis was performed using SAS v. 9·2 (SAS Institute Inc., USA). The proportion of false-positive reports of meningococcal disease in hospital data was estimated by examining those children with an additional hospital/mortality record after the original hospital admission. Comparisons between subjects were performed using χ 2 and two-sample t tests. Capture–recapture analysis was performed using the Chapman estimator [Reference Chapman19] for notification and hospitalization data. Capture–recapture estimation using three datasets (notification, hospitalization, mortality) was performed with the genmod procedure using a log-linear Poisson regression model [Reference Orton, Rickard and Gabella20].
The remoteness of residence was defined by using the mean Accessibility/Remoteness Index of Australia (ARIA) value corresponding to a subject's statistical local area of residence [21] in NSW. Mean ARIAs <0·20 were considered to be major cities, between 0·20 and 2·40 were inner regional areas, and >2·40 were outer regional and more remote areas. A subject's sex and Aboriginal/Torres Strait Islander status were taken from the first hospital record where meningococcal disease was diagnosed.
Approximate yearly incidence rates between 2000 and 2007 (per 100 000 child population) were calculated by dividing the number of notified cases of meningococcal disease in NSW-resident children by the NSW child population (0–14 years). This incidence rate uses all available data to maximize the number of recorded cases of this rare disease. To calculate an incidence rate/100 000 child-years at risk, the dataset was limited to only those children resident and born in NSW (i.e. those with a perinatal data collection record) to give both the total child-years at risk of meningococcal exposure and the number of notified cases of meningococcal disease in this subset of children. Incidence rates were presented both as cases/100 000 child population and cases/100 000 child-years at risk to facilitate comparisons with other published works. Incidence rates were adjusted for the estimated false-positive rate in hospital data, the degree of under-ascertainment found through capture–recapture analysis, and the estimated interstate migration rate for children.
RESULTS
A total of 595 cases of meningococcal disease were notified between July 2000 and December 2007. No subject had more than one notification of meningococcal disease. The majority of meningococcal disease cases [343 (66%) of 522 laboratory-confirmed cases] were serogroup B. Ninety cases (17%) were classified as serogroup NOS, 66 cases (13%) were serogroup C, 17 cases (3%) were ‘not typable’, and six cases (1%) were one of the less common serogroups (groups W, Y or 29E).
There were 684 children who had one or more episodes of hospitalization coded as meningococcal disease between July 2000 and December 2007, of whom 466 (68%) also had a notification of meningococcal disease. ABS mortality data recorded 26 children who died from meningococcal disease. A total of 813 cases were identified as having meningococcal disease in at least one of the three datasets. Figure 1 illustrates the overlap between cases of reported meningococcal disease identified in the three datasets.
All 813 cases recorded in notification, hospitalization, and/or mortality data as meningococcal disease in children, July 2000–December 2007. (Figure is not to scale.)
Of the 466 children both hospitalized and notified with meningococcal disease, 58% were male, 12% were resident in outer regional or more remote areas of NSW, and 7% were Aboriginal or Torres Strait Islander.
Fatal cases of meningococcal disease
All 26 deaths with a cause of death listed as meningococcal disease had a notification of meningococcal disease, although four of these children did not have an associated hospital admission (see Fig. 1). In two of these non-hospitalized fatal cases, the causes of death included Waterhouse–Friderichsen syndrome, or fulminant meningococcal disease, the most severe presentation of the disease [Reference Barquet22]. It is possible that some of these children died before they could be admitted to hospital.
Estimating false-positive disease reporting by observing multiple clinical records for single subjects
Just over one-third (38%, 257/684) of cases with a hospital diagnosis of meningococcal disease had at least one further clinical record (all had hospital admission/s and 24 cases had a death record) which followed the initial meningococcal hospital admission within a period of 24 h. The 466 notified cases with either single (n = 323) or multiple (n = 143, including 22 deaths) clinical records were considered ‘proven cases’ of meningococcal disease (group 1). ‘False-positive cases’ were not notified, had multiple clinical records, and lacked a diagnosis of meningococcal disease on the final record (group 2, n = 103 including two deaths). These represented 90% (103/114) of all non-notified cases with multiple clinical records. Non-notified subjects with a single clinical record (n = 104) or with multiple clinical records where the final clinical record included a diagnosis of meningococcal disease (n = 11) could not have their status determined and were classified as ‘unknown’ (group 3, n = 115). Such subjects (90% of whom had a single clinical admission) may represent either proven cases who were not notified, or cases who did not have a subsequent clinical record and the opportunity to revise the diagnosis. Table 1 demonstrates significant differences between group 3 ‘unknown’ cases and either groups 1 ‘proven’ or 2 ‘false-positive’ subjects. Group 3 subjects were not significantly different from either groups 1 or 2 subjects by sex, remoteness of residence, time in intensive care, or Aboriginal or Torres Strait islander status.
Significant differences between group 1 (proven), group 2 (false-positive) and group 3 (unknown) subjects
Group 1 ‘proven’ cases: those notified as having meningococcal disease. May include subjects with single or multiple clinical records.
Group 2 ‘false-positive’ cases: non-notified subjects who had been discharged without a meningococcal diagnosis recorded on their final clinical record. Only includes subjects with multiple clinical records.
Group 3 ‘unknown’ cases: non-notified subjects who had been discharged with a meningococcal diagnosis on their final (or only) clinical record. May include subjects with either single or multiple clinical records.
* Analysis of all categorical variables is based on the first hospital separation for the illness episode. Continuous variables are summed over all hospital separations in the illness episode.
† One missing value in group 1.
A single NSW hospital accounted for 40% (41/103) of all group 2 false-positive cases of meningococcal disease during the study period. All other NSW hospitals recorded no more than eight group 2 cases each.
The most frequent primary diagnoses in the final clinical record for those where meningococcal disease was excluded by the final clinical record (group 2, n = 103) were: viral infection, unspecified (n = 40), unspecified viral infection with skin and mucous membrane lesions (n = 7), acute upper respiratory infection, unspecified (n = 5) and fever, unspecified (n = 5).
Using capture–recapture methods to estimate under-ascertainment of meningococcal disease
Capture–recapture methods assume that subjects are equally likely to be ‘captured’ by the individual datasets [Reference Hook and Regal23], which would require that the same definition of meningococcal disease in hospital and notification data. As this assumption was not met, we adjusted our capture–recapture estimates for the greater proportion of false-positive cases in hospital data.
Assuming that the ‘false-positive’ rate of 90% that we identified in non-notified cases with multiple hospital records was applicable to all non-notified hospitalized cases (n = 218), we estimated that up to 22 of these may actually be ‘true’ (confirmed or probable) cases of meningococcal disease. Table 2 shows capture–recapture estimates [Reference Chapman19] for the true number of cases of meningococcal disease, after adjustment for the estimated false-positive rate in the hospitalization data.
Cases ‘captured’ and ‘not captured’ by notification and hospitalization data, after adjustment for estimated false-positive reporting in hospital data
Non-italicized numbers have been observed in the data. Italicized numbers have been calculated based on assumptions described in the text. N was calculated using the Chapman estimator (here showing the upper limit):
$$N = \displaystyle{\left( {n_N + 1} \right)\left( {n_A + 1} \right) \over \left( {n_N,A} + 1 \right)}\, - \,1; N = \displaystyle{{596 \times 489} \over {467}} - 1; N = 623.$$
* An estimated upper limit of 22 true cases of meningococcal disease which were admitted to hospital, but not recorded in notification data
Capture–recapture methods estimated that up to six cases of meningococcal disease in children aged between 0 and 14 years might not have been reported in either hospitalization or notification data between July 2000 and December 2007. The capture–recapture estimate for the total number of cases of meningococcal disease in this period ranged from 595 to 623.
For the subset of NSW-born and -resident children included in Table 3, capture–recapture methods estimated that up to four cases of meningococcal disease may not have been captured by either hospitalization or notification data in the study period, and the total number of cases ranged from 435 and 455.
Cases ‘captured’ and ‘not captured’ by notification and hospitalization data, after adjustment for estimated false-positive reporting in hospital data in the subset of NSW-born and -resident children
Non-italicized numbers have been observed in the data. Italicized numbers have been calculated based on assumptions described in the text. N was calculated using the Chapman estimator.
Incidence rates
The incidence of notified meningococcal disease calculated using the total annual population of NSW-resident children aged 0–14 years averaged 5·6 (95% CI 4.4–7.0) cases/100 000 child population over the study period. Annual incidence declined from 7·7 (95% CI 6·3–9·4) cases/100 000 child population in 2001 to 4·1 (95% CI 3·1–5·3) cases/100 000 child population in 2007. Annual incidence of serogroup B also decreased from 3·4 (95% CI 2·5–4·6) cases/100 000 child population to 2·9 (95% CI 2·0–3·9) cases/100 000 child population over the same time period. Throughout the years, the highest incidence of meningococcal disease occurred in children aged <5 years (Fig. 2).
Crude incidence of meningococcal disease in NSW children by year.
In the subset of NSW-born and -resident children the linked datasets contributed 8 456 477 child-years at risk. Using the capture–recapture estimates for the number of cases in hospital and notification data, adjusted for false-positive reporting in hospital data (435–455 cases) yielded an incidence density estimate of between 5·1 and 5·4 cases of meningococcal disease/100 000 child-years at risk in NSW children aged 0–14 years over the study period.
The likely impact of interstate migration was estimated indirectly, using previously published figures. Between 2001–2002 and 2010–2011, NSW had an average net population loss of 23 349 persons/year [24], and between 5% and 8% of these people (about 1500/year) were aged between 0–14 years [25], so that about 9750 child-years were lost to interstate migration over the study period. Subtracting this figure from the total child-years at risk left the incidence density estimate unchanged. Other international population-based estimates of the incidence of meningococcal disease in children are given in Table 4.
International estimates for the incidence of meningococcal disease in selected populations
DISCUSSION
Due to its rarity, diversity of presentations, and difficulties in diagnosis, we tested the possibility that cases of meningococcal disease were under-ascertained in existing NSW recording systems using capture–recapture analysis of linked administrative datasets. We used linked notification, hospitalization and mortality data to investigate potential under- and over-ascertainment of cases of meningococcal disease in these recording systems, and to produce population-based estimates of the incidence of meningococcal disease. We identified high levels of capture in notification data of fatal cases (100%), and around 15% of children hospitalized with an initial diagnosis of meningococcal disease were ‘false-positive’ cases in whom the diagnosis was not subsequently confirmed. Incidence rates calculated using adjusted capture–recapture estimates and person-time denominators derived from linked data did not differ appreciably from those calculated using notification data only and ABS population estimates.
Completeness of notification data
Our capture–recapture methods, using hospital and notification data and correcting for estimated false-positive reporting in hospital data, demonstrated that most confirmed and probable cases of paediatric meningococcal disease in NSW are reported to the Notifiable Conditions Information Management System. An estimated maximum of four cases of meningococcal disease in NSW-born and -resident children may have gone unreported in hospital or notification data between July 2000 and December 2007, less than one case/year.
This is an improvement on meningococcal disease surveillance conducted between 1990 and 1995 in Victoria. That study estimated seven cases unidentified in children aged <2 years, but complete ascertainment for children aged 10–14 years [Reference Robinson26]. During this period, no culture was available for 38% of cases [Reference Robinson26] compared to our study in the following decade, where more accurate laboratory diagnosis methods using PCR were in use, and many notifications were coming directly from the laboratories themselves.
False-positive reporting in hospital data
One of the assumptions of capture–recapture techniques is that databases have the same disease reporting criteria (which implies similar false-positive reporting rates) [Reference Hook and Regal23]. Hospital and notification data do not meet this assumption: hospital data records ‘possible’ cases in addition to ‘confirmed’ and ‘probable’ cases of meningococcal disease, so hospital data has a greater proportion of false-positive cases than notification data. While other studies have used clinical review to identify false-positive diagnoses [Reference Breen11], this is not possible using de-identified data linkage files. Instead, we examined the subset of non-notified patients with multiple clinical records and estimated that at least 90% were false-positive cases of meningococcal disease. This is a novel approach to estimating the degree of false-positive reporting in linked data.
To examine our assumption that group 3 ‘unknown’ cases more closely resembles group 2 ‘false-positive’ cases, and was less similar to group 1 ‘proven’ cases of meningococcal disease, we refer to Table 1. In the clinical variables of both age group and time on mechanical ventilation, there was a significant difference between groups 1 and 3 (but not between groups 2 and 3), which would seem to imply that group 3 unknown cases were not cases of meningococcal disease. The significant difference in ages between groups 1 and 3 could demonstrate the greater incidence of meningococcal disease in young children (as per Fig. 2) in proven (group 1), but not unknown (group 3) cases.
In the clinical variables of presenting symptoms and length of stay, a gradient of severity was observed, but group 3 was significantly different to both groups 1 and 2, providing no further evidence to reveal the identity of group 3 subjects. For both of these variables, group 1 (proven) cases were the most severe, group 3 (unknown) cases had intermediate severity, and group 2 (false-positive) cases were less severe.
Two of the significant differences between groups 2 and 3 (the role of the presenting hospital and the year of presentation) were not clinical in nature. These differences could reflect changing policies on the referral of, and the lower capacity of smaller district hospitals to deal with cases of suspected meningococcal disease. Hospital-level variation between groups 2 and 3 can been seen clearly in the proportion of false-positive group 2 cases by hospital. A single NSW district hospital accounted for 40% of all non-notified, false-positive cases of meningococcal disease in NSW. The emergency department of this particular district hospital saw over 10 000 paediatric patients per year, yet had no paediatric unit/ward at the time and all paediatric admissions were transferred to a larger nearby hospital for assessment and treatment [27]. These final non-meningococcal diagnoses were predominately unspecified viral infections, which is consistent with false-positive meningococcal disease reporting.
Dependence of hospital and notification data
Capture–recapture methods also assume that the relevant datasets are independent, so that subjects ‘captured’ in one dataset are not more or less likely to appear in another dataset [Reference Hook and Regal23]. In our sample, there is a positive dependence between hospitalization and notification data, in that hospitalized cases are more likely to be notified. Nearly half (47%) of first notifications for meningococcal disease came from a hospital, followed by 30% from laboratories, 7% from a doctor, and 16% from other sources. This positive dependence is likely to bias the capture–recapture estimate towards underestimation [Reference Hook and Regal23]. We did attempt to adjust for this dependence between the databases by using a third (mortality) database and a log-linear capture–recapture model [Reference Orton, Rickard and Gabella20]. However, as all fatal cases of meningococcal disease were captured by notification data, mortality data contributed no additional information to the estimate.
Incidence of meningococcal disease
The average incidence of notified meningococcal disease in NSW children was 5·6 (95% CI 4·4–7·0) cases/100 000 child population over the study period, and this incidence decreased over the study period. In 2007, there were 4·1 cases (95% CI 3·1–5·3) of meningococcal disease and 2·9 cases (95% CI 2·0–3·9) of meningococcal group B/100 000 child population. Using our capture–recapture estimate and the subset of NSW-born and -resident children, there was an adjusted incidence density of between 5·1 and 5·4 cases of meningococcal disease/100 000 child-years at risk (July 2000–December 2007), which was very similar to our crude estimate from notified cases only. While not directly measured in this study, levels of interstate child migration did not have a significant impact on these estimates. These incidence rates are comparable to others from Australia and Denmark where meningococcal disease notification systems and/or record linkage were used [Reference Chiu4, Reference Olesch and Knight28–Reference Menzies30], but much lower than incidence rates in the UK calculated using only hospital diagnoses [Reference Heyderman31] or Spanish rates which included all meningococcal diagnoses, whether identified from hospital or laboratory data [Reference Barquet22] (see Table 4).
CONCLUSIONS
Obtaining accurate measures of disease incidence, including rare diseases with a difficult diagnosis such as meningococcal disease, is important so that we can best assess the impact and cost efficacy of interventions such as immunization. Our investigation shows that linked data can be used to estimate the proportion of false-positive disease reporting in hospital data where clinical review is not possible. Capture–recapture estimates show that paediatric meningococcal disease was well recorded in NSW notification data, and we obtained an internationally comparable estimate of the incidence of meningococcal disease in NSW children.
ACKNOWLEDGEMENTS
The authors acknowledge Brett Archer for comments on an earlier version of this manuscript. This project was funded by the National Health and Medical Research Council (grant no. 573 122).
DECLARATION OF INTEREST
None.
