- Research article
- Open Access
- Open Peer Review
Heterogeneity in regional notification patterns and its impact on aggregate national case notification data: the example of measles in Italy
© Williams et al; licensee BioMed Central Ltd. 2003
- Received: 18 December 2002
- Accepted: 18 July 2003
- Published: 18 July 2003
A monthly time series of measles case notifications exists for Italy from 1949 onwards, although its usefulness is seriously undermined by extensive under-reporting which varies strikingly between regions, giving rise to the possibility of significant distortions in epidemic patterns seen in aggregated national data.
A corrected national time series is calculated using an algorithm based upon the approximate equality between births and measles cases; under-reporting estimates are presented for each Italian region, and poor levels of reporting in Southern Italy are confirmed.
Although an order of magnitude larger, despite great heterogeneity between regions in under-reporting and in epidemic patterns, the shape of the corrected national time series remains close to that of the aggregated uncorrected data. This suggests such aggregate data may be quite robust to great heterogeneity in reporting and epidemic patterns at the regional level. The corrected data set maintains an epidemic pattern distinct from that of England and Wales.
- Measle Case
- Uncorrected Data
- Case Notification
- Notification Data
Highly infectious diseases such as measles necessitate vaccination of a very high proportion of the population in a sustained and effective manner in order to achieve control. Therefore national policy makers must possess data which reflect as closely as possible the true scale and pattern of infection, thus allowing vaccination policy to be tuned to achieve maximum impact.
Optimisation of vaccination programme design can be greatly facilitated through insights gained from the mathematical modelling of measles transmission dynamics . Such modelling can shed light on reasons for observed patterns of infection in both pre-vaccination and post-vaccination eras and upon likely effects of the continuation of existing policies. However, modelling work must be based upon sound data if full confidence is to be placed in results and, since epidemiology of specific infectious diseases may differ according to geographic area, the availability of good local data is of great importance. Seroepidemiological surveys provide reliable information about patterns of experience of infection, e.g. Salmaso et al , but when considering historic data, and in the absence of continuing large scale representative programmes of serological surveillance, reliance must be placed upon case notification data for information about patterns of infection over time.
To date, many of the modelling studies have focused on measles with much important analysis having been based on a relatively small number of long data series. These provide a valuable epidemiological resource, but their value relies heavily upon the consistency and the reliability of the relevant systems of case reporting. One European example is measles case notification data from England & Wales, providing raw material for much work in this field [3–6], and widely considered to be of good quality although, even here, concerns have been raised about under-reporting, and misreporting of measles cases remains a danger ; also notification efficacy has been found to be more accurate during periods of high incidence than when infection is rare . Elsewhere in Europe, availability and quality of case notification data has varied widely between and within countries . Case based surveillance systems have been employed in the majority of European countries (including Italy and the UK), although case definitions have varied until recently and some countries have also required additional laboratory confirmation or epidemiological linkage; in other European countries sole reliance has been placed upon sentinel systems [9, 10]. More reliable data should result from the European surveillance network (EUVAC-NET) recently constituted with the aim of providing a joint data base with a uniform measles disease classification [9, 10]. Additionally, the European Sero-Epidemiology Network (ESEN), was set up in 1996 to gather age-stratified immunity data for measles (and other diseases) in Italy, the UK, Denmark, France, Germany, and the Netherlands, although, being based on residual sera, the data does not form a random cross-sectional survey [9, 11].
We have here attempted to overcome these inconsistencies by making use of the observed fact that, in the pre-vaccination era, nearly all were infected by measles well before reaching adulthood . Thus, for each region, disregarding migration, there should be an approximate correspondence over time between regional numbers of births and numbers of measles cases. A simple procedure is employed here making use of this correspondence to estimate the degree of under-reporting. These estimates are used to revise regional notifications data in order to provide a corrected aggregate national data set; results are compared with estimates of national case notifications from measles mortality data (for the purposes of further comparison a similar procedure was employed for the national data set of measles cases for England & Wales).
Although in essence this approach to correction of data is not novel, being an extension of that of Clarkson & Fine  (see also  and ), we believe that what is new is its systematic application to regional datasets to provide corrected aggregate national data and that this approach does provide a much sounder basis for the epidemiological analysis of national data sets. Subsequent publications will discuss the time series analysis of this data set, an investigation of the corrected national age distribution of cases and its variation over time.
Cases 1000 live births-1
Measles notifications for Italy by region.
Friuli Venezia G.
England & Wales
Table 1 makes clear the wide variation in numbers of case notifications between regions in both pre- and post-vaccination eras. At its greatest, during the pre-vaccination period, this reaches an order of magnitude difference between Emilia Romagna in the north (209.3 notifications 1000 live births-1) and Campania in the south (22.6 notifications 1000 live births-1). If data for the provinces corresponding to the 4 largest cities (population > 106) is taken into account, the difference is even greater with Campania's regional capital of Naples providing only 16.0 notifications 1000 live births-1. For half of the regions there is also clear evidence of a decline in the number of notifications per 1000 births during the pre-vaccination period (not shown). These contrasts are maintained on a wider geographic scale with, for example, figures of 108.9 and 30.4 notifications 1000 live births-1 in North and South Italy for 1949–76 and, for 1987–96, 76.9 and 34.0 respectively. In contrast, the national notification rate of 572.6 notifications 1000 live births-1 for England & Wales for the pre-vaccination period 1949–66 is some 8 times greater than that for Italy (73.1) in the comparable period 1949–76.
Smoothing of births and case notifications
The trends of smoothed case notifications data for the Italian regions (see Additional file: 1 and Additional file: 2) were more varied than those of births, although generally slightly declining towards the start of vaccination in the mid-1970's, in some cases from a broad peak or plateau and in others with occasional sharper peaks or other fluctuations (in contrast smoothing of the national pre-vaccination notifications data for England & Wales showed a more or less constant number of cases). After the start of vaccination in the mid-1970's, some Italian regions saw a decrease in notifications, but others, strangely, an increase (perhaps occasioned by a heightened awareness of the need for monitoring cases following the introduction of vaccination); yet others showed a decrease followed by increase.
Estimates by region of Italy of percentage of measles cases notified for periods specified.
Friuli Venezia G.
Effect of different weighting systems on estimates of under-reporting.
% cases reported
Catalytic model using force of infection (FOI) from Italian seroprevalence data
Catalytic model using FOI from case notifications data
Catalytic model using EURO FOI (see text)
Age distribution of cases 1971–76
Simple moving average
Corrected case notification data for Italy
The degree of under-reporting seen in Italian regional notification data of the pre-vaccination era is in some instances, strikingly large, so that corrected aggregate national data for Italy is an order of magnitude greater than for the uncorrected data. Nevertheless, although they vary widely between regions, shapes, location and trends of adjusted and unadjusted national data are very similar.
Case reports and measles deaths
Although the proportion of cases notified may be poor, it seems likely that deaths arising from measles cases would be more reliably reported, so that a relatively high ratio of deaths to cases might be expected in the circumstances described. Figure 3 confirms this expectation by comparing notified deaths and cases for the period 1958–77 with those observed in England & Wales over similar and earlier timescales . It may be argued that the higher Italian 'deaths to cases' ratios may in part result from poor living conditions or standards of health in certain regions (in this regard an indication of the ratio from one developing country (Peru) is also included for comparison). Some support for this notion is provided by a comparison of infant mortality rates for Italy (52.7 and 35.6 1000-1 live births in 1951–60 and 1961–70 respectively ) with those of England & Wales (29.1 and 17.7 1000-1 live births in 1950–2 and 1970–2 respectively) and Peru (47.48 1000-1 live births in 1995 ) but it is also likely that the high Italian 'deaths to cases' ratio also reflect the very poor levels of case reporting illustrated above  (the term 'case fatality rate' is deliberately avoided because of the low level of case reporting).
Under-reporting in the vaccination era
Age bias in reporting
Overall, though corrected age-aggregated notification data for Italy are certainly much closer to true numbers of cases, the issue raised by Edmunds et al  of age bias in reporting of measles cases in Italy remains to be considered (this stands as a counter-example to the observation of Fine and Clarkson  that the probability of a case of measles being reported appears higher among young children than adolescents and adults). If there is indeed a substantial increase in under-reporting with decreasing age, there would be an overestimation of the average age of infection and underestimation of the force of infection in the youngest age groups. This may in turn impact on health policy both directly and, indirectly, through the effect on parameter estimation for modelling work informing health policy, and, of course, vaccination programmes themselves will influence the average age of infection and so modify the effect of reporting age-bias. It is therefore of great importance that further investigations be undertaken to address this issue of age bias.
Clearly there is a need to improve overall levels of case notification within the regions of Italy. So far, no formal studies to investigate the reasons for under-reporting have been conducted. Nevertheless, the experience of a study conducted with a sentinel network of primary care pediatricians shows that simplicity of reporting and regular feed-back of results can greatly improve case notification. A recent evaluation of varicella under-reporting in Italian children and adolescents, showed also that case notification was more complete in primary-school-aged children compared with that for other age-groups , probably because medical certificates for readmission to school are required after certain infectious diseases, including measles and varicella. Such certificates are usually provided by public health officials who may be more aware than general practitioners of the relevance of infectious disease notifications.
Nevertheless, it is clear that further research will need to be undertaken if we are to understand fully the reasons for such poor levels of reporting and if sound strategies for optimising the impact of vaccination are to be developed. However, it is difficult to foresee a mechanism for overcoming the apparently dramatic extent of under-reporting in some regions that would not depend upon a substantial investment in monitoring systems.
Despite qualifications outlined above, it is believed that the adoption of the procedure described here provides, by eliminating potential distortions arising from wide variation in regional reporting levels, both i) a national time series of measles data, and ii) a national age distribution of cases, of improved reliability which will prove a useful basic resource to aid further understanding of the epidemiology of measles infection, both within Italy and, when similar substantial regional heterogeneity in notification levels become apparent, in the wider context.
Sources of data
Monthly measles case notification data were provided by the Istituto Nazionale di Statistica (ISTAT) for the period from the first available year, 1949, to 1996. These show substantial heterogeneity between regions. Previously published data on annual births in Italy and measles deaths in Italy and England & Wales were also used.
In Italy there is a pattern of mixed private and public sector delivery of vaccination and measles vaccination is classed as 'Recommended' rather than 'Compulsory' (as is also the case for mumps and rubella). Nationally vaccination against measles was first made available towards the end of 1976 and became officially recommended in 1979. There is little data on vaccine uptake but initially it is believed to have been low  and, for the first 10 years at least, to have remained at disappointingly low levels , perhaps exacerbated by the fact that only a proportion of local health units provide vaccine free of charge.
Estimation of degree of under-reporting
As note above, prior to widespread commencement of measles vaccination, most individuals were infected by measles before adulthood . Clearly because of the cyclic epidemic pattern, variation in the age at which individuals are infected, and fluctuations and trends in birth numbers, there could not be an exact correspondence between births and cases. Nevertheless, ignoring migrational flows (see below) and assuming endemic equilibrium, during the pre-vaccination period numbers of measles cases over time would have been approximately equivalent to numbers of births, providing a basis for the estimation of the degree of under-reporting .
One approach to such estimation would be to select a suitable value for the force of infection (FOI: the per capita yearly incidence of infection in the susceptible proportion of the population) and, by using a simple catalytic model , to estimate the equilibrium age distribution of cases and hence distribution of the birth years of those infected in an average year. Although perhaps theoretically sound, this has two main drawbacks. First, it is quite demanding in terms of births data. Second, it presumes availability of some "suitable" FOI (a reasonable choice used here could be the EURO FOI estimated by Edmunds et al  for a wide range of European countries). This assumption is clearly not neutral (nor even tautological), as it implies imposing the structure of the hypothesised FOI upon the very data from which subsequently an estimation of the true FOI would be made. Instead we seek here a correction tool based on minimal assumptions, to minimise possible biases introduced by the correcting algorithm.
In pre-vaccination Europe around half of all measles cases would be expected to occur in the first 4–7 years of life , with numbers of cases at older ages becoming more and more widely distributed with increasing age. In these circumstances it was felt that estimation of under reporting using a moving average to smooth yearly birth data would be more than adequate (see below for a theoretical justification). In the event, values obtained by this means proved sufficiently close to those derived using the catalytic modelling approach above to additionally justify its use in this instance on grounds of simplicity, greater transparency and ease of application as a general method. A births curve smoothed by averaging over some appropriately delayed period of years , should encapsulate much of the magnitude and trend of pre-vaccination measles cases. Therefore the ratio of the moving averages of births to that of notifications should provide a suitably robust estimate of the degree of under-notification. Here moving averages of both births and cases for each region were calculated as:
Where x t denotes the original time series of births or case notifications, as appropriate, n +1 the number of terms included in the moving average and y t the moving average series. The estimated level of under-reporting, U t , was therefore simply:
where C t = y t(cases) and B t = y t(births). The lag, m, between averages for births and those of cases is intended to take into account the cohorts supplying the greater proportion of cases occurring in any given year [5, 15]. As an example, in the present instance where a lag of 4 years was chosen (the same as used by Clarkson & Fine , U1957 is computed from birth numbers for the years 1949–57 and case notifications for the years 1953–1961.
Using births as estimators for cases: theoretical justification
We consider the simplest SIR compartmental model  describing infection in a stationary homogeneously mixed population in absence of vaccination:
where X(t), Y(t), Z(t) represent numbers of susceptible, infective and recovered/immune individuals at time t, B(t) births per unit time, λ (t) = βY(t) the force of infection (FOI), μ the mortality rate (constant for simplicity), and ν the (constant) recovery rate. The total population N = X + Y + Z is assumed stationary with B = μN. At equilibrium numbers of births (B) and cases (C) respectively are B = (μ + λ)X and C = λ X, so that:C = B - μX.
A prediction of equations (i) is that the susceptible fraction at equilibrium, X/N, is the reciprocal of R 0 (the basic reproduction ratio of the infection . Thus: C = B - μN/R 0 = B (1-(1/R 0)). Therefore for very high values of R 0 such as for pre-vaccination measles, with a sufficiently low mortality rate, C/B≈1. This relation is basically the consequence of the fact that the age window over which the forces of mortality and infection operate are separated. It is worth noting that the basic relation between births and cases holds in more general circumstances. Suppose there is a steady oscillation of (i) around its long term equilibrium, as is typical of periodically forced SIR and SEIR models. In this case the relation C/B≈1 remains correct provided we consider average values of cases and births over the appropriate time period. More generally if the model is non-stationary, provided state variables are bounded in the long term, the equilibrium condition (time derivatives of the state variables are zero) may be replaced by the weaker condition that their long term averages are zero: , so that E(C) = E(B) - μE(X) indicating that the basic relation still holds in a broader sense, i.e. if yearly numbers of cases and births are replaced by their averages over a sufficiently long period.
The previous relations are preserved when age-structure is explicitly introduced. Consider a SIR model with chronological age structure  in a stationary population with no additional mortality arising from disease
where X, Y, Z, λ, μ are as in equation (i) with the added dimension of age, a.
With a stationary population B(t) = B, and assuming equilibrium, the total number of cases of infection per unit time is:
where p(a), Λ (a) denote respectively survival to death and infection functions. Under Type I mortality (μ (a) = 0, a <L, μ (a) = ∞, a ≥ L), commonly used to approximate mortality in industrialised developed countries:
where F(L) = 1 - Λ (L) is the fraction who have experienced the disease by age L. Since for measles in the pre-vaccination era F(L) is usually very close to 100%, again the relation C ≈ B holds. In the special case of homogeneous mixing by age one finds in particular C = λ X = B(1 - e -λL ).
Using the approximate relation A = 1/λ (holding under Type I mortality ) one gets the further relations showing that yearly numbers of cases at equilibrium are quite close to yearly numbers of births for large R 0 values. Under Type II mortality (μ (a) = μ) one recovers the relation C = B (1 - (1/R 0)) found for (i).
A possibility offered by (iii) for estimating the degree of under notification of measles cases under more general circumstances appears if one explicitly introduces time, showing that the total number of cases is a weighted average of past births
the weights being given by the (normalised) density of infection: G(a) = Λ (a)λ (a)/(1-Λ (L)); this is the catalytic approach mentioned in the text. As long as births do not fluctuate wildly, cases can thus be reconstructed from births provided a suitable force of infection is assumed. The further simplified approach used in this paper follows by considering time averages of (v) over a generic interval (t0,t1):
By the mean value theorem a value ζ exists such that
(vii) suggests that a moving average of births, delayed by a suitable choice of ζ, may be used to reconstruct the time series of present (moving averaged) cases. The literature (and references therein) suggests that possible shapes of the age densities of infection for measles in pre-vaccination regimes are sufficiently well behaved so that the choice of ζ is easy. Compared to (v), (vii) has the advantage that assumptions about the full shape of the force of infection function are unnecessary.
Finally we point out that the relationship between cases and births is robust to changes in basic model assumptions, extending to more general models such as SEIR's. Moreover, if the population is exponentially increasing rather than stationary (typical of developing countries) it holds in a more general form involving discounted births rather than actual births.
The effects of migration
Force of infection values estimated from Italian case notifications (ICN-UR) and from those from a number of European countries (EURO)
Values of the force of infection (%/year)
This exercise suggests substantial movements of susceptible individuals: e.g. during 1951–61 for typical industrial "sink" regions, such as Piemonte and Liguria, taking immigration into account expected case numbers were up to 15% larger (under ICN FOI) than those expected by ignoring migration. Despite this, the impact of migration on estimated levels of under-reporting is rather limited. Only in a few regions are there absolute variations (|Uincl.migration/Uexcl.migration|) of at most 2% (usually much less) with respect to what is predicted by the basic cases-births relation. This broadly confirms the robustness of the simplified approach followed in the paper. Full details of results on estimation of under-reporting in presence of migration are avaliable on http://statmat.ec.unipi.it/manfredi.html.
This paper was written within the research project Progetto Nazionale "Epidemiologia delle Vaccinazioni in Italia". John Williams was funded under a project grant from the Italian Ministry of University and Scientific Research. We thank Donatella Mandolini, John Edmunds, Alberto Tozzi, and Antonino Bella for their valuable comments and suggestions. Usual disclaimers apply.
- Anderson RM, May RM: Infectious diseases of humans: dynamics and control. 1991, Oxford, OUPGoogle Scholar
- Salmaso S, Gabutti G, Rota MC, Giordano C, Penna C, Mandolini D, Crovari P: Pattern of susceptibility to measles in Italy. Bulletin of the World Health Organization. 2000, 78: 950-955.PubMedPubMed CentralGoogle Scholar
- Fine PEM, Clarkson JA: Measles in England and Wales: I, an analysis of factors underlying seasonal patterns. International Journal of Epidemiology. 1982, 11: 5-14.View ArticlePubMedGoogle Scholar
- Anderson RM, Grenfell BT, May RM: Oscillatory fluctuations in the incidence of infectious disease and the impact of vaccination: Time series analysis. Journal of Hygiene. 1984, 93: 587-608.View ArticlePubMedPubMed CentralGoogle Scholar
- Grenfell BT, Anderson RM: The estimation of age-related rates of infection from case notifications and serological data. Journal of Hygiene (Cambridge). 1985, 95: 419-436.View ArticleGoogle Scholar
- Finkenstadt B, Grenfell BT: Empirical determinants of measles metapopulation dynamics in England and Wales. Proc. R. Soc. London B. 1998, 265: 211-220.View ArticleGoogle Scholar
- Ramsay M, Brugha R,, Brown D: Surveillance of measles in England and Wales: Implications of a national saliva testing programme. Bulletin of the World Health Organization. 1997, 75: 515-521.PubMedPubMed CentralGoogle Scholar
- Clarkson JA, Fine PEM: The efficiency of measles and pertussis notification in England & Wales. International Journal of Epidemiology. 1985, 14: 153-168.View ArticlePubMedGoogle Scholar
- Butler AR: Measles, mumps and rubella (MMR) immunisation programmes in Europe: analyses of the impact on the incidence of measles based on mathematical models of viral transmission dynamics. Zoology. 2002, Oxford, Oxford, 402-Google Scholar
- Glismann S, Rønne T, Schmidt J-E: The EUVAC-NET survey: national measles surveillance systems in the EU, Switzerland, Norway and Iceland. Eurosurveillance. 2001, 6: 105-110.PubMedGoogle Scholar
- Osborne K, Weinberg J, Miller E: The European Sero-Epidemiology Network. Eurosurveillance. 1997, 2: 29-31.PubMedGoogle Scholar
- Santoro R, Ruggeri FM, Battaglia M, Rapicetta M, Grandolfo ME, Annesi I, Cortellessa CM: Measles Epidemiology in Italy. International Journal of Epidemiology. 1984, 13: 201-209.View ArticlePubMedGoogle Scholar
- Finkenstadt BF, Grenfell BT: Time series modelling of childhood diseases: a dynamical systems approach. Appl Statist. 2000, 449: 187-205.Google Scholar
- Ramsey M, Gay N, Miller E, Rush M, White J, Morgan-Capner P, Brown D: The epidemiology of measles in England and Wales: rationale for the 1994 national vaccination campaign. CDR Review. 1994, 4: R141-R146.Google Scholar
- Edmunds W John, Gay Nigel J, Kretzchmar M, Pebody RG, Wachmann H: The pre-vaccination epidemiology of measles, mumps and rubella in Europe: Implications for modelling studies. Epidemiology & Infection. 2000, 125: 635-650.View ArticleGoogle Scholar
- Sniadack DH, Moscoso B, Aguilar R, Heath J, Bellini W, Chiu MC: Measles epidemiology and outbreak response immunization in a rural community in Peru. Bulletin of the World Health Organization. 1999, 77: 545-552.PubMedPubMed CentralGoogle Scholar
- Livi-Bacci M, Breschi B: Italian fertility - an historical account. Journal of Family History. 1990, 15: 385-View ArticleGoogle Scholar
- US Census Bureau: Statistical Abstract of the United States: 1993. 1993, Washington DC, US Government Printing Office, 113Google Scholar
- Blackburn N, Schoub B, O'Connell K: Reliability of the clinical surveillance criteria for measles diagnosis. Bulletin of the World Health Organization. 2000, 78: 861-PubMedPubMed CentralGoogle Scholar
- Ciofi degli Atti ML, Salmaso S, Bella A, Arigliani R, Gangemi M, Brusoni G, Tozzi AE, Group SPES-Study: Paediatric sentinel surveillance of vaccine-preventable diseases in Italy. Pediatric Infectious Disease Journal. 2002, 21: 763-769.View ArticlePubMedGoogle Scholar
- Ciofi degli Atti ML, Rota MC, Mandolini D, Bella A, Gabutti G, Crovari P, Salmaso Stefania: Assessment of varicella underreporting in Italy. Epidemiology and Infection. 2002, 128: 479-484.View ArticlePubMedPubMed CentralGoogle Scholar
- ICONA working group: ICONA: national infant vaccination coverage. 1998, Istituto Superiore di SanitàGoogle Scholar
- The pre-publication history for this paper can be accessed here:http://0-www.biomedcentral.com.brum.beds.ac.uk/1471-2458/3/23/prepub
This article is published under license to BioMed Central Ltd. This is an Open Access article: verbatim copying and redistribution of this article are permitted in all media for any purpose, provided this notice is preserved along with the article's original URL.