The data collection and methods are described in detail in a previous article [18], but will be presented briefly here.

In Denmark, all patient contacts with hospitals, both inpatient and outpatient, are recorded in the National Patient Registry with a primary diagnosis and any optional secondary diagnoses [19]. It includes data from all patient contacts, and thus, it is representative of all patients diagnosed with ‘neck injuries’ seen in a hospital. More specifically, data containing the ICD diagnostic code of ‘cervical-column distortion’ (S134) were extracted from the registry for this study. Data were available throughout the 12-year observation period from 1998–2009. The pre- and post periods covered up to 3 years before and after the diagnosis year, but patients were only included in the calculations if they had at least 1 year prior to or post the diagnosis year in the observation period. That is, the patients included in the co-morbidity analysis were diagnosed between 1999 and 2008. For the recording of mortality, all available years within the 12-year period after the individual injuries were included. Thus, all individual data on hospital co-diagnoses can be traced retrospectively and/or prospectively.

All patients with a first-time diagnosis of a neck injury were included, irrespective of which type of accident had led to that diagnosis, and irrespective of whether or not the diagnosis was a primary or secondary diagnosis. There is some underestimation of the national number of patients with neck injuries, because those who contact the primary care sector, in contrast to the hospital sector, are recorded as having had contact, but without necessarily a diagnosis being recorded.

To assess what neck trauma this diagnosis actually covered, we sampled all patient records with this primary diagnosis from the emergency department of Glostrup University Hospital from 2 Jan 2013 to 10 June 2014. We then selected all those born on the first day of any month (irrespective of the month or year of their birth), then added all those born on the second day of any month, and so on, until 100 consecutive patients had been selected. We then categorised the trauma mechanism of the injuries of these 100 people into: whiplash from rear-end collision/whiplash from frontal collision/whiplash with direction not indicated/whiplash-like, not due to a car accident but neck injury sustained through a head trauma/other trauma.

By examining the National Patient Registry, we identified all patients whose first diagnosis of neck injury was in the period from 1998 to 2009. This was chosen because the first year that valid hospital cost data were available was 1998 and the last was 2009, and because we had used this interval for the other studies assessing health care system costs [2, 19]. Then, using data from the Civil Registration System Statistics Denmark Database (which includes information about social factors, marital and cohabiting status, incomes, pensions), we randomly selected matched control subjects with the same age, sex, and marital status as each individual patient, but without a registered diagnosis of neck injury. Parity of socio-economic status was achieved by selecting controls from each patient’s county of residence. A patient-to-control ratio of 1:4 was used to optimise the representativeness of the controls. Data from patients and control subjects who could not be identified in the Coherent Social Statistics database were excluded from the sample. All of these characteristics in both groups were successfully matched.

The patients and control subjects were followed over the entire study period. Specifically, we recorded health data for up to 3 years before, and up to 3 years after, the year of injury. Thus, for individuals with an injury e.g. in year 4 (=2001) in the 12-year period, both the injured and the matched controls were tested for co-diagnoses for the years 1–3 (i.e. before the injury) as well as the years 5–7 (after the injury). Those with a recorded injury in the years 2–3 as well as 10–11 contributed less before and after data respectively. Those with injuries that occurred at the beginning of the 12-year period and who could therefore not contribute ‘before data’, and those with injuries that occurred at the end of the 12-year period who could not contribute ‘after data’, were not included in the calculation of co-morbidity. However, all mortality data from 2009 were included.

We found that those with a registered neck injury showed a co-morbidity rate that was approximately 1½ times higher than that of the controls, as assessed from the recorded diagnoses, both prior to and after the neck injury diagnosis.

This and our earlier study with the same sample appear to challenge the value of usual case-control studies, because what is often seen during follow-ups after an accident or other sickness may not have been caused entirely by that accident or sickness. This means that the accident or sickness itself is probably not the total pathology observed during follow ups, which instead consists of the injury in combination with pre-existing frailty and/or sickness behaviour. Thus, if the study had been carried out with the injury as the starting point, the findings would have been highly different from what was demonstrated from our case-control study.

For future case-control studies, it is strongly recommended to find relevant pre-values for the included individuals from their health-care records or public health registries. If this is not possible, which is often the case, an estimation of selection bias, for example, using the one described by Geneletti et al. [22], should be carried out. This is a mathematical model based on simulations and logistic regression, designed to solve/reduce this problem in case-control studies.

To date, criticisms of case-control and similar studies focus particularly on ‘selection bias’ as a great source of error. This was described in the review by Lee et al. [23], who reviewed non-randomised, controlled studies, that started with the onset of a disease. Out of 408 such studies on various psychiatric disorders from journals with an impact factor >3, the controls were individually matched to the cases in only 30 % of studies. In the remainder, the controls were convenience samples of local students/coworkers, people recruited via advertisements or similar. They also included other methodological pitfalls, such as having inadequate sample sizes, etc.

A characteristic morbidity pattern was qualitatively noted: both prior to and after the injury the most prevalent diagnostic groups were ‘code 19: injuries etc’; ‘code 13: musculoskeletal’; and ‘code 18: miscellaneous abnormal clinical and laboratory findings’ (Tables 3 and 4).

The big question is why, on average, people prior to their neck injury already had 1½ times more hospital-diagnosed co-morbidities. It is not likely that those whiplash accidents stemming from rear-end collisions represent people with a predisposition to being more frail before the accident. Several alternative explanations exist.

One explanation could be that some of the cases had a sustained low threshold for care-seeking and therefore had had more diagnoses registered [24]. Along this line, it could be that some of the controls who did have a minor to moderate rear-end whiplash trauma, were less likely to contact the health care system as compared with the cases in this study.

Another possible explanation is that some of the cases were prone to thrill-seeking behaviour and therefore more likely to be involved in accidents. An additional argument could be put forward that some of the accidents were the result of the injured people being generally more clumsy in their movements. However, these human behavioural characteristics may not explain all of the at least 46 % of the accidents where the person was hit by a car from behind, and these represent approximately two-thirds of those accidents where a direction of impact was reported. Those with ‘no direction indicated’ (Table 2) most likely reflect patient interview inaccuracy or poor journal entry practice. Therefore, we might assume that it is likely that two-thirds of those with no direction indicated (i.e. 8/12) were also rear-end, and thus our best estimate is that approximately 54 % of all injuries were rear-end accidents. It is however very unlikely that all ‘whiplash-like accidents’ (16 %) and ‘frontal collisions’ (24 % + possibly some of those ‘no-direction indicated’) should have been caused by thrill-seeking or clumsy behaviour.

The fact that ‘abnormal clinical and laboratory findings ..’ were also over-represented in the injured group may be explained by more frequent care-seeking visits for any health condition, and therefore abnormal findings more often being noted.

That all the other diagnoses were approximately 1½ times more frequently recorded among cases than controls, may reflect that a certain proportion of cases were more frail both before and after the accident. For example, a person with pre-existing musculoskeletal problems may be bothered more after a certain accident than someone without such a history.

However, this trend may not apply to those with rear-end collision, and it seems very unlikely that it should explain all the other accident types.

None of these potential explanations is sufficient to fully explain our findings, and it is likely that all components may play a role in some way. Even the previously reported increased health-related costs prior to injury [2] may also reflect a mixture of predisposition to frailty and a general low-threshold for health-care seeking.

Furthermore, medico-legal and economic issues may be related to an increased likelihood of an individual receiving a hospital diagnosis and engaging in more health-care seeking. Regarding post-whiplash earnings, Leth-Petersen et al [25] found that those not granted compensation returned to pre-injury income, whereas those granted compensation had chronic Whiplash Associated Disorders more often, and remained at a lower income post-injury compared with pre-injury. Also their patient records showed more frequent health care visits pre-injury. Regarding this socio-economic aspect and our findings, it might also be that those with pre-existing health and/or social problems might be more acutely in need of economic support or benefits after an accident, and therefore might be more prone to care-seeking than those more well off.

What also complicates the conclusions is the observation that the culturally-based conceptual framework of a health condition may highly impact the choice of diagnosis in cases with non-specific illnesses. A complicated example was observed before the fusion of East and West Germany, where back pain as a cause of sick leave was reported as more predominant in the west than in the east, which evened out during the following years after the re-union [26]. A similar example is that the diagnosis ‘repetitive strain injury’ almost disappeared during the 1980s after evidence that the condition did not relate to ergonomic conditions nearly as much as previously believed [27]. Also, an additional issue of varying thresholds for pain chronicity in individuals – with or without a neck injury – was reported in a recent study that used functional cerebral MRI to suggest that even at an early stage of LBP, cerebral white-matter low-myelin composition could be related to an increased likelihood of developing chronicity [28].

We could speculate that our findings may be explained by one or more of these factors: (i) that cases have a tendency towards poor pain-coping behaviour with a higher focus on sickness and a lower threshold for visiting a doctor; (ii) that a low threshold for care-seeking for milder sickness episodes more often resulted in cases receiving the various co-morbidity diagnoses reported here; (iii) that a proportion of the cases really were more frail and may explain some of those injuries not related to a rear-end collision; and (iv) that some injuries may stem from thrill-seeking behaviour or being more clumsy, resulting in more injuries and therefore also more co-diagnoses.

The important finding was that co-morbidities that occur in the years following a neck injury may not be due entirely to the accident, but represent an already existing predisposition to poor health and care-seeking: (i) an already existing lower threshold of health care-seeking among those registered with a neck-injury diagnosis which would explain some of the neck-injury co-morbidities, (ii) a proportion of the neck injured individuals, other than those having had a rear-end-collision whiplash trauma, who are generally more biologically frail prior to the injury due to a poorer health constitution, and (iii) a proportion of the neck injuries occurring in individuals with thrill-seeking behaviour or generally being more clumsy.

An additional and important implication of these findings is that results from matched-control studies should be cautiously interpreted, because health status following a studied event may reflect an existing predisposition and not be entirely a consequence of the event.