Here, we discuss the advantages and disadvantages of different methods for identifying UE in the surveillance pyramid and compare MFs resulting from those methods. MFs show considerable between-country and -disease variation which may reflect true differences in reporting and ascertainment rates. However, study design (which often depends on the disease, data type, data quality and availability of resources) and hence the method used to estimate UE likely accounts for the presented within-country variability of MFs. It remains difficult to select the most appropriate study (and corresponding MFs) as there are limitations associated with each. CBS that are representative of the whole population are often favoured for approximating UE, UA, and UR but the quality of MFs derived depends highly on the study design; CRS are very good at estimating UR in the surveillance pyramid but some reported cases may remain undetected and UA is not considered; and RTS and serological surveys also estimate UE effectively but the many associated limitations must be realised. In addition, an important limitation is the uncertainty that surrounds estimates which is relevant to all study types. Many of the studies from the literature review do not report uncertainty for MFs which gives a false impression that we are sure that these point estimates are correct. In fact, a great level of uncertainty is expected
. Since the factors contributing to UE are multiplicative (Figure
1B), they explode rapidly and the resulting MF can be very large (this is clearly observed in pyramid reconstruction studies which aim to account and correct for UE at each incremental step of the surveillance chain). Therefore, even small degrees of uncertainty in measuring individual components of UE can lead to wide ranges in incidence estimates and MFs. To estimate predictive intervals, uncertainty can be modelled by incorporating, for example, either uniform or pert probability distributions (rather than fixed point estimates), and using techniques such as Monte Carlo simulations
[4, 50, 80, 84].
One systematic way to decide on the best method for estimating UE or choosing MFs is to use the Delphi method or expert consensus. In the next phase of the BCoDE-project, internal ECDC experts with final input from the consortium and other external experts will create lists of the most appropriate country- and disease-specific MFs for 32 IDs. In general we assume that the most appropriate MFs should be disease-, country-, age-, and sex-specific because underestimation rates are disproportionately distributed between diseases, countries with differing surveillance systems and reporting procedures, and between demographic groups. While our literature review returned only one result with age- or sex-stratification of MFs, there are other studies that provide age- (at least for given age bands)
[12, 21, 34, 116–119] and sex-specific
[117, 118] MFs for general gastroenteritis and diarrhoea (i.e. unspecified pathogen). However, there remains a paucity of age- and sex-specific MFs in the literature.
Where no MF exists, it is (under certain conditions) possible to 'borrow’ or extrapolate from a disease of similar epidemiology or from the same disease in a country with a similar surveillance system, or likewise apply the same MF to a group of diseases (as demonstrated by Mead et al.). However, it must be acknowledged that the base value of “cases reported” (Figure
1A and B) that we seek to adjust for UE by applying an appropriate MF, may not always capture the same proportion of infections that have occurred or provide comparable information of disease incidence estimates for different diseases or countries. Therefore, borrowing MFs (particularly from different countries and especially from different disease groups (e.g. taking MFs for STIs and applying to gastroenteric disease data)) is not a favoured method owing to the inherent heterogeneity of national surveillance systems in terms of population covered, test sensitivity and specificity, the source of data (physician, laboratory, hospital or other) and surveillance type (whether compulsory versus voluntary reporting of positive results or cases, comprehensive versus sentinel, active versus passive surveillance, case-based versus aggregated reporting
Here we did not address UE in the mortality reporting chain. Similar to the surveillance pyramid for morbidity data, the tip of mortality pyramid represents the cases correctly reported. The wide base of the pyramid contains data of all deaths including those that are ascertained and those that are not. Whilst it is expected that in a European setting under-ascertainment of deaths is rare (if not irrelevant), underreporting or over-reporting of mortality events due to certain diseases or conditions is not. The number of deaths may be well reported, but there is considerable misclassification of the cause of death. This misclassification may be deliberate in countries without nationalised healthcare such as the United States, where reimbursement by private insurance may be related to the ICD code used for the primary cause of death. Elsewhere, there may be other reasons to misclassify the cause of death, such as government targets and pressure to reduce the number of deaths due to a certain cause. In addition, often lacking are additional details relating to underlying conditions and sequelae that an individual died with (e.g. secondary and tertiary causes) but not necessarily of (i.e. the primary cause of death)
[13, 121, 122]. For example, chronic conditions with infectious causes (e.g. liver cirrhosis) are often not counted as sequelae deaths and therefore the surveillance system may underestimate the long-term burden due to the infection that led to the sequelae.