Humans can get many different co-infections, but treatment guidelines only exist for a few specific combinations (e.g., HIV and hepatitis C). Co-infection morbidity has also been studied within certain cohorts (e.g., 5–16 year olds, ), and is often reported to be worse than single infections . However, the occurrence of co-infection in death across age and sex cohorts has, to our knowledge, never been studied before. Our results indicate that (i) co-infection death may be more common in early adulthood, but it is not known whether younger adults are more susceptible to co-infection per se, or more susceptible to fatal co-infection. We also found that (ii) pairs of infections with strong positive association on death certificates tended to co-occur more often than those with strong negative associations. This suggests that medical care of severely ill patients with some co-infections can be problematic. Finally, (iii) co-occurrence on death certificates was positively related to biologically similarity. Better understanding of these biological interactions may help efforts to predict and combat co-infection mortality. We discuss the factors that may contribute to these patterns, before considering implications for treatment, limitations of the data, and future research needs.
The early-to-mid adulthood peak in co-infection death contrasts with theories that the immune response declines in old age , and with non-infectious diseases where comorbidities increase with age . This could be explained by individuals being more susceptible to death from one infection in old age, either because they are frailer as their bodies deteriorate through oxidative damage , or the infection coincides with non-infectious causes of death that are more common with age, like cancer . Alternatively, young adults are more prone to severe immunopathologies following infection: critically ill patients with influenza A(H1N1) tended to be 20–30 years old , and the added physiological stress of co-infection might make death more likely. Another possibility is that more effort is made to find infections in critically ill young adults than for older patients. We are not aware of evidence that biased medical practices also contribute alongside the physiological factors mentioned above, but this is a possibility that could be examined further.
Reasons for males being at higher risk of infection than females include behaviours that put them at greater risk of infection, or physiological reasons, such as sex hormones, that make them more susceptible to severe disease once infected [7, 28]. Our data do not enable us to distinguish which of these mechanisms may have played a role. If males undertake riskier behaviour, have higher testosterone in early adulthood, or are less likely to visit the doctor when ill this may explain why the sex difference appears around the peak of the distribution (Fig. 1).
Our results suggest that co-infection treatment guidelines could be based on synergistic interactions between infections. Most possible pairs of infections co-occurred on death certificates at a frequency expected from their occurrence alone. We suggest that the unassociated pairs of infections could be excluded from efforts seeking to identify severe co-infections.
Around 1 in 20 possible pairs were associated and tended to co-occur more often than expected. Positively associated pairs of reported co-infections included: mycobacteria and HIV, viral hepatitis co-infection, and cytomegalovirus and pneumocystis. While these similar pairings were often reported together, associations were context dependent; they were negatively associated with other infections, including mycobacteria and infectious bloody diarrhoea, pneumocystis and sequelae of tuberculosis, and viral hepatitis and Zoster virus infection. The direction of association is therefore not consistent for the same infection, and so treatment guidelines should not be based solely on the identity of one constituent infection. Whether the relatively weak correlations are clinically meaningful remains a debatable point, but on a population scale, across hundreds of thousands of deaths, the results suggest that it may be important to public health and worthy of further investigation. The biological similarity of associated pairs could be an important consideration when assessing the potential severity of a given co-infection.
Data quality and limitations
Studies based on reported data must consider potential biases. In our dataset there may be underreporting of co-infection death on death certificates if infectious disease was undetected, wrongly deemed not to have contributed to death, or were not reported using multiple codes. Poor reporting of causes of death was a problem in the UK in the 1990s . There have since been legal and educational reforms , and death certificate data have been audited by the Center for Disease Control and the Office for National Statistics. Using multiple infectious causes as indicators of co-infection probably underestimates the true number of co-infection deaths. One could hypothesise that certain types of infections, such as those detected by the same test, with similar tropism, of high severity, might be more likely to be diagnosed. However, from death certificates alone we are unable to examine whether behaviour or diagnostic techniques may have played a role. We have no evidence of systematic bias that could have generated the patterns we found, but we encourage further broad scale analyses of co-infection to help establish the key factors of the individual and their infections that can best guide treatment. Our conclusions are robust to the complexity of model fitted (Additional file 1: Supplementary Information S1), measure of association used (Additional file 1: Supplementary Information S2), country and method for analysing biological similarity (Additional file 1: Supplementary Information S3 and S4), ambiguity in ICD-10 codes (Additional file 1: Supplementary Information S5), and inpatient status (Additional file 1: Supplementary Information S6). Therefore, we are confident that we describe genuine patterns.
Other limitations to the secondary data available include: an inability to distinguish certain pathogens within the ICD-10 disease codes, severity of disease not necessarily corresponding with both infections being of the same timescale, and the age categories reported being somewhat arbitrary and not matching physiological changes like puberty.
Causes of death are associated with various factors including healthcare, socioeconomic status, family structure, geography, behaviour, physiology, or infectious dose. Determining what factors affect causes of death using national observational data alone is difficult. Co-infection death needs to be assessed in other time periods and countries.
The patterns we described could be attributed to biological interactions, or an artefact of the relative prevalence of the infections among at-risk populations. To disentangle the two we need data on co-infection prevalence. While we have some evidence that the number of reported infection deaths is not correlated with reported infections in England and Wales (Additional file 1: Supplementary Information S7), this was only for a subset of infections, and we could not find data on prevalence for most co-infections in our dataset.