Discussion
Principal Findings
From 1991 to 2012, more than one in 10 initial antibiotic monotherapies for four common infections were associated with treatment failure, as defined by a prespecified set of decision criteria. For upper and lower respiratory tract infection, skin and soft tissue infection, and acute otitis media, the overall failure rate increased from 13.9% to 15.4% over this period, and from 14.6% to 15.3% averaged over the earliest and latest five year periods. With 1991 as the reference year, by 2012 there was an adjusted increase in failure rates of 12% for the four classes overall. Most of the increase was from 2000, when community antibiotic prescribing, which had been falling in the second half of the 1990s, plateaued and then once again began rising.
The rate of increase in antibiotic treatment failure was less prominent with the most commonly prescribed antibiotics and in those recommended for first line treatment, such as broad spectrum penicillins (amoxicillin) and the macrolides (clarithromycin and erythromycin). There were notable increases in failure rates for some antibiotics that are usually not recommended as first line treatments for the conditions under study, such as trimethoprim, the cephalosporins, and the quinolones. Such drugs, however, might have been prescribed for more severely ill and frail patients who had recently been prescribed a first line drug or from whom resistant organisms had been previously isolated.
Strengths and Limitations of the Study
The analysis of antibiotic treatment failure within primary care as a measure of effectiveness has not been attempted before on this scale. We analysed almost 11 million antibiotic monotherapy treatments for four common infection groups in research quality patients.
As the Clinical Practice Research Datalink collates data from routine practice, there are inevitably missing and erroneous data, coding imperfections, and variations in medical practice. We militated against this to some extent by applying the organisation’s data quality metrics. Of the patients excluded from the study on the basis of the research quality of their records, most (78%) were classed by the Clinical Practice Research Datalink as unacceptable on the basis of their temporary registration status, which could have resulted in the under-representation of certain disadvantaged groups with higher mobility and, possibly, greater burdens of illness. The poor quality of data associated with temporary registration, with respect to patient history and follow-up, however, made this decision necessary. The number of antibiotic treatments consequently excluded (1.7%) is unlikely to affect our findings.
Exposure to the antibiotic treatments of interest was inferred from the prescription records, which could only be considered as an intention to treat. We could not determine whether the patient actually redeemed the prescription or whether the antibiotic was taken correctly. The direct effects (as opposed to drug resistance effects) of non-compliance on the failure rates observed here would probably be to inflate the number of treatment failures identified. Any trend towards increased compliance, possibly caused by improved patient and practitioner awareness, might therefore reduce failure rates over time. We do not know whether increasing use of delayed prescribing influenced our findings.
Indications could not be determined by our algorithm for some 60% of the monotherapies, largely because the diagnostic codes associated with these monotherapies described symptoms such as cough that could not be reliably allocated to an infection class or represented conditions that fell outside our selected infection classes. This affected some antibiotics more than others; for example, 91% of metronidazole monotherapies had undeterminable indications compared with 45% of phenoxymethylpenicillin monotherapies. The sets of diagnostic codes selected to identify indications were, of necessity, limited to those in which the site and nature of the infection were clear. We believe that the missing indications should not change the overall findings as far as the most commonly used antibiotics are concerned. The Appendix shows a summary of the phenotypic characteristics of those excluded.
The study’s classification of upper respiratory and skin and soft tissue infections might have been too broad for meaningful comparison, and future analysis would subdivide upper respiratory tract infection into acute sinusitis and laryngopharyngitis, the better to differentiate their pathology, and would distinguish between acne vulgaris—for which chronic repeat prescriptions are often issued—and other skin and soft tissue infections.
Antibiotic Treatment Failure and Bacterial Resistance
Our long term data characterising antibiotic treatment failure in primary care, derived from one of the few sources available to investigate such trends, showed that there was evidence of an increase in failure rates. Rates could be influenced by several factors, including changes in host factors (such as adherence to treatment, the social determinants of health, or reduced immunity associated with the older profile of the sample in more recent years), antibiotic related factors (such as changing dose and treatment duration), and pathogen related factors (such as changing virulence and antimicrobial resistance). The failure rates we observed might be lower than those observed in hospital care because infections in primary care are more often viral or self limiting. Signals of actual bacterial resistance might therefore not become apparent against a background of unnecessary antibiotic use. The possible underestimation underlines the necessity of making better, diagnosis guided, responsible decisions about antibiotic prescription in daily practice.
Although prescriptions are automatically recorded in the Clinical Practice Research Datalink with regard to product and quantity, precise dose instructions are not always entered explicitly. For this reason, we did not investigate treatment dose, although a preliminary exploration suggested that antibiotic doses might have increased over time. This warrants further investigation, with possible revision of the definition of antibiotic treatment failure to include dose intensification. When treatment failure increases despite higher antibiotic doses, this will be of great concern.
Implications of Study
This study characterises antibiotic prescribing and treatment failure in primary care in the UK. Internationally, however, antibiotic prescribing practices in the community vary markedly. Antibiotic consumption has been most extensively characterised in the European Union, where, in 2011, this ranged from 11.4 defined daily doses per 1000 population in the Netherlands to 35.1 defined daily doses in Greece. The UK was ranked 17th highest of the 29 countries surveyed, with prescribing, at 18.8 defined daily doses per 1000 population, being more than 60% higher than in the Netherlands. Such variation at a country level—the product of many complex factors, including drug regulatory, educational, and cultural differences—has been correlated with resistant invasive infections in hospitals. The results obtained here for the UK cannot therefore be directly applied elsewhere without consideration of local factors. It remains, however, that the UK is one of few countries that has the coverage and continuity of community based data to enable a study such as this to be carried out at a patient level.
A detailed summary of the current state of antibiotic resistance throughout the world was published recently by the Lancet Infectious Diseases Commission. While the evidence from this commission borders on the alarming, the consensus seems to be that there is still a chance of dealing with this public health threat.
We have shown that in primary care, where most antibiotics are prescribed, and in a developed country, the impact of increasing antibiotic treatment failure was not as great as we had anticipated based on evidence reported from hospital settings with more severe infections. Another interpretation of these data, however, would be that failure rates have indeed increased over this period and that the observed rates are in reality higher because inappropriate or unnecessary prescribing could have attenuated the failure rates. Given the lack of new antibiotic classes on the horizon, increases in failure rates are troubling.
For the recommended first line treatment antibiotics, such as broad spectrum penicillins and the macrolides, failure rates were relatively stable over an extended period of more than 20 years, while those for other antibiotics, often not considered first line for the indications we studied, increased in some notable cases. The highest rates were seen with lower respiratory tract infections, where affected patients tended to be older and less healthy.
Conclusions
We do not know whether the increases in antibiotic treatment failure we identified represent a phenomenon that will resolve or whether this is an early indication of a more dramatic and worrying process. Nevertheless, the finding that treatment failure was associated with more than one in 10 initial antibiotic treatments in primary care represents a considerable burden on patients and on the healthcare systems. Not only should rates of antibiotic resistance continue to be monitored and acted on, trends in failure rates should also be closely scrutinised. Similar analyses should be carried out for other countries that have varying rates of antibiotic resistance to further explore the association between treatment failure and antibiotic resistance.
Our data suggest that primary care physicians could play a central role in helping to contain rises in antibiotic treatment failure by managing patients’ expectations and carefully considering whether each prescription is justified; once the decision is made to prescribe an antibiotic, the choice should follow current guidelines regarding first line drugs.