Sensitivity and specificity of molecular methods for detecting markers of antimalarial drug resistance in clinical samples of Plasmodium falciparum: a systematic review


Malaria doesn’t need much of an introduction. It’s big (91 countries with ongoing transmission, millions of cases per year), and bad (hundreds of thousands of deaths each year, two thirds which are of children under 5 years)(World Health Organization, 2017b).

Neither does antimicrobial resistance, first commonly recognised in the 1940s (penicillin-resistant staphylococcus, streptomycin-resistant tuberculosis) and now widespread to treatments for tuberculosis, HIV/AIDS, malaria, sexually transmitted diseases, urinary tract infections, pneumonia, blood-stream infections and food poisoning (World Health Organization, 2015).

The good news is that malaria is still preventable and treatable. Some parasites are resistant to one or more antimalarial drugs (but not all) and resistance is developing and spreading. Surveillance of resistance is essential to inform malaria prevention and treatment policies; most surveillance is carried out by monitoring genetic markers (specific mutations in one of a few genes) of resistance (World Health Organization, 2017a).

There is no scientific consensus on the best methods to detect these genetic markers; different labs and different individuals within the same labs use different methods to detect these markers. Using substandard methods to detect these markers of resistance will be generating less data, and less accurate data, and misinform antimalarial prevention and treatment policies – leading to more cases and deaths from malaria.

I have worked in malaria in different labs for a number of years, and have been frustrated to find out that we do not know how best to monitor resistance or have an estimate of how inaccurate our existing data might be. I wanted to answer these questions!

So I undertook this systematic review (the first of its kind) to collate and compare estimates of test accuracy for detection of antimalarial resistance markers in malaria (Burrow, 2017).

I structured my research question using the PIRT framework:


blood samples from patients with, or suspected of having malaria (Patient group)

how accurate is

any molecular method (Index text)

compared to

any other molecular method (Reference test)

for detecting

antimalarial drug resistance markers (Target condition)



I included all diagnostic accuracy studies that examined at least two molecular methods for detecting selected markers of antimalarial resistance in blood samples from patients diagnosed with, or suspected of having malaria. I searched MEDLINE, EMBASE, BIOSIS, and Science Citation Index. I used QUADAS-2 to evaluate methodological quality. I calculated sensitivity and specificity for each and synthesized and compared the results through narrative.


I had a horrifying 27,575 search results, of which 36 studies were potentially eligible. Of these, 15 were eligible and extra information successfully sought for another three. That’s a total of 18 studies, examining 13 index tests, against five reference tests, for their accuracy in detecting 24 molecular markers. The PRISMA flow diagram and overall risk of bias are in Figures 1 and 2.

Figure 1. PRISMA Flow Diagram

Figure 1

Figure 2. Overall Risk of Bias and Applicability Assessment

Figure 2

I have a lot of results tables, so you’ll have to read my dissertation if you want to know the sensitivities and specificities I found. Perhaps more interesting is what I didn’t find…


Half of the 36 studies were not eligible for inclusion as data allowing calculations of sensitivity and specificity were not made available. What difference would these results have made? This could be evidence of publication bias (abstracts were published but not full length papers) and/or of poor reporting practises. No studies were registered, prospectively or retrospectively, so the number of studies reported could not be compared to the number of studies planned. Of the 18 eligible studies, none were reported according to STARD guidelines; most papers were published before the development of the guidelines. However, they still failed to report many details that are difficult to define as anything other than essential (for example, results).

Most studies were underpowered for detection of at least one molecular marker.

No exact replication of studies was done, and very few studies replicated either the combination of reference and index tests or any one test for a specific molecular marker.

The design of studies, often the selection of patients and samples in particular, made it seem as though diagnostic accuracy studies had been added on to other types of studies as an afterthought. Studies were almost all either designed in a way that made them susceptible to large biases, or the design was not reported and biases and applicability were unclear.

Estimates of sensitivity and specificity were calculated, but confidence intervals were wide (partly owing to the above issues) and the direction, size, and effect of biases is mostly unclear.

In addition, most studies were carried out in non-endemic countries using equipment that is not available in most surveillance situations.


I don’t think my findings should change clinical practise, but they should change research practise. I think researchers need to be deliberately asking questions about test accuracy, focussing on methods that are being used and will be used in endemic countries, planning high-quality studies to answer those specific questions, using sample size calculations, prospectively registering their studies, reporting studies according to STARD, responding to requests for further information, replicating these studies, and all of this led by, and in partnership with, researchers in malaria-endemic countries.

Systematic Review Registration: PROSPERO CRD42017056249


BURROW, R. 2017. Sensitivity and specificity of molecular methods for detecting markers of antimalarial drug resistance in clinical samples of Plasmodium falciparum: a systematic review. MSc, University of Oxford.

WORLD HEALTH ORGANIZATION 2015. Global Action Plan on Antimicrobial Resistance. In: WORLD HEALTH ORGANIZATION (ed.). World Health Organization.

WORLD HEALTH ORGANIZATION 2017a. Status report on artemisinin and ACT resistance (April 2017). In: WORLD HEALTH ORGANIZATION (ed.) Status report on artemisinin and ACT resistance. World Health Organization.

WORLD HEALTH ORGANIZATION 2017b. World Malaria Report 2017. In: WORLD HEALTH ORGANIZATION (ed.) World Malaria Report. World Health Organization.



I’m working on my MSc dissertation (Sensitivity and specificity of molecular methods for detecting markers of antimalarial resistance in clinical samples of Plasmodium falciparum: a systematic review) and I’ve been trying to track down eligible studies.

All the studies should be easily found in bibliographic databases, and all the data I need should be in the published papers, right?

Of the 35 potentially eligible studies:

  • 17 report too little data for me to be able to extract the results I need.

What would the results of my review be if I could include this data? Maybe leaving out nearly 50% of eligible studies which report data from over 50% of participants isn’t ideal and might affect the validity of my results?

I thought trial registries might be useful for finding out more.

Of 35 potentially eligible studies:

  • None were registered, either prospectively or retrospectively.

(One study had an associated study of adherence to rapid diagnostic tests registered – results not provided).

So I have no idea if study protocols were changed, had missing outcomes, new outcomes…

Trial registries were not helpful for checking details of studies I know exist. How many studies are there that I don’t know exist?

Are my systematic review results going to be bollocks due to all the different biases introduced by crappy study designs and ad hoc changes, unpublished studies, outcome switching, and missing data from published studies?

And how can I find out, if I can’t ever know how much data I’m missing or how biased the data I have is?

It should be so simple..

REGISTER AND REPORT – diagnostic test accuracy studies can be registered at any of the registries listed in Table 1 of this great paper, and should be registered somewhere prospectively. They should be reported according to STARD.

Even using the resources I have free and easy access to – the Bodleian, the British Library, and kind friends at plenty of other institutions with awesome libraries, I still couldn’t get all the papers I might have wanted. So what is everyone else supposed to do?

OPEN ACCESS – often funded by public money, often work done by public employees, time and samples and personal data always donated by members of the public – make OUR data available to us, the people you’re doing it for/with.

Fair disclosure, we didn’t register the one study on which I’m a co-author. Or report it following STARD. I am going to register it retrospectively. And we did report enough data in the published paper for inclusion in my systematic review.

Bugs AND bollocks

Stemming from one research paper is a google* of total wank.

*now a collective noun



Minus points to The Sun for worst pun-biggest misleading stat combination



So who has covered themselves in glory? From the top hits, pretty much everyone. NPR, WebMD, The Sacramento Bee (?), CBS, Time, the BBC, the Huffington Post, The Sun

Many of these pieces contained small sections discussing the limitations of the research paper, and clarifying the difference between correlation and causation. BUT. They spent WAY MORE space speculating about possible mechanisms that would explain a causal link. The overwhelming feeling reading each of these articles is that there is little doubt that pubic hair removal causes an increased risk of STIs.


I found one well-written article at Lifehacker which made it clear through that this was an association and…


…(excited squeak) ACTUALLY LINKED TO THE RESEARCH PAPER! Full points.



I’m not a fan of the research paper itself.


Observational studies are important. Observational studies that are well designed to investigate their question of interest are much more important. Obviously. This is not one of those.

In a study of a possible association with, or cause of, STIs, would you:

  1. Want to know about safer sex and condom use? These researchers didn’t.
  2. Assume that people are open and honest about sex and STIs? These researchers did.
  3. Assume that people can accurately remember their lifetime history of STIs and pubic hair removal? These researchers did.




Younger people might remove more hair, more often than older people. They might be having sex more often with more people, more recently. They are doing this in an era with higher prevalence and transmission of STIs.

People who are lucky enough to be having lots of sex or sex with lots of people might be more inclined to remove more pubic hair more often. Ask any woman about hair removal. EVERYONE KNOWS you don’t make a real effort unless someone is going to notice. Shave your legs in the winter in the absence of a new(ish) naked friend? NO WAY. Bikini line unless you’re going to the beach? NEVER.


Science = having sex means a higher risk of STDs.

Common sense = having sex means higher chance of removing body hair.


For the benefit of the Huffington Post:


People removing pubic hair are having more sex with more people, and the more sex with more people is giving them more STIs. That’s why.


Probably. I’m going to need a really interesting RCT to be sure.

Some unconnected and enjoyable advice…


Magic Bacteria I

I try to be logical about which health charities I support (those that make the greatest difference to the most people) and which research excites me (useful, high quality) but my brain snags on Alzheimer’s.

Alzheimer’s is a big deal. Dementia (including Alzheimer’s) currently affects approximately 48 million people. It can affect “memory, thinking, orientation, comprehension, calculation, learning capacity, language, and judgement”, which is basically everything that makes you, you. There’s no cure, no treatment, and we don’t really know why some people get it and others don’t.

So a bit more clarity and the hope of any treatment would be a wonderful thing.




The Mail gets in a little bit of fear-based selling on the side…

A quick look, and you might be considering taking probiotic supplements.

So why does this report make me grumpy? Not because I used critical appraisal tools to carefully consider the paper. I got bored half-way through. Instead, here are some pet peeves.

A 12-week study is NOT LONG ENOUGH. People can have Alzheimer’s for over a decade. I need to know what happens to people who have Alzheimer’s for more than 12 weeks.

60 patients is NOT ENOUGH PARTICIPANTS, especially as 8 of them died before the study finished. The trialists managed to get some statistically significant stats – but having lost 13% of their participants, perhaps you would expect some variation between baseline and end of study results? How did they know 60 people was enough to detect the difference they hypothesised? How applicable are these results of these 60 people to the other 48 million?

15 different biomarkers, WHO CARES. Might be useful for mechanistic proof of concept. But are they outcomes relevant to patients? Or have they be stuck in post-hoc to ramp up the number of significant outcomes? Outcome switching.

Mini-mental state examination (MMSE) started at 8.47 (±1.10) and decreased to 8.00 (±1.08) in the control group. MMSE started at 8.67 (±1.44) and increased to 10.57 (±1.64) in the probiotic group. The numbers are statistically significantly different. But are the patient’s symptoms clinically significantly different? NHS Choices says probably not (they already thought of the things I wrote). So WHAT IS THE BENEFIT FOR PEOPLE WITH ALZHEIMER’S?

This trial does not show a benefit (or show lack of benefit) of probiotics in reducing symptoms of Alzheimer’s – because the trial hasn’t been well enough designed. And the reporting was crappy. THIS MAKES ME GRUMPY.

What’s the evidence: Do chickens prevent malaria and Zika infections?

It’s been four weeks since this paper was published, but I’ve been on holiday and received a lot of insect bites, so my interest in a chicken mosquito-deterrent hasn’t waned.

The Telegraph had the headline:

“Suspending a chicken over your bed could protect against Zika virus and malaria”.

They also had some fairly confident remarks from one of the authors about the size of the effect

“can cut populations by up to 95 per cent throughout an entire house, so it’s very efficient”

and possible utility of their tested compounds

“I think it should [prevent Zika]. We haven’t tested it on other mosquitoes but there are lots of varieties which won’t feed on chickens and so would be repelled.”

Malaria kills hundreds of thousands of people each year. Zika is disabling and afflicting thousands this year. Headlines and quotes like this generate a huge amount of hope. But is it false hope? What’s the evidence?


Photo by Muhammad Madhi Karim

This paper included many different useful and interesting pieces of work. The researchers:

  • collected data on the populations of human and domestic animal species in three villages
  • collected blood-fed mosquitoes from 10 houses and 5 pit shelters in three villages
  • identified mosquito species and source of last blood meal
  • collected headspace samples from 5 individuals from each species in 1 village
  • reared an Anopheles arabiensis colony
  • analysed chemicals present in the headspace of different species
  • measured electrophysiological response of antennae to samples and isolated compounds
  • counted mosquitoes captured in CDC suction traps baited with control solvent, synthetic compounds or a live chicken


This is a huge amount of work, so I’m going to focus on the question that seems to be of most interest:


Can chickens, or compounds extracted from chickens, reduce malaria or Zika infections in humans better than current mosquito repellents?


If this is the question of interest (although we may be more interested in whether we prevent deaths, or disability, rather than reducing the number of cases) then the key evidence is that generated in the final point. From the other aspects of this paper we can see that they started off well. They identified villages with mosquitoes. They identified the species of mosquito that would be present, and that these mosquitoes might be likely to bites humans indoors. They identified compounds that showed activity in a laboratory setting. Then they carried out a randomised controlled trial to see whether these compounds would reduce the number of mosquitoes captured in a trap near sleeping humans.


They found that compared to solvent only, the four identifiable chicken compounds, two host compounds, and a live chicken, significantly reduced the number of mosquitoes captured in a suction trap. This sounds really promising – but there are problems.


Surrogate outcome

They’re not measuring how many cases of malaria or Zika were prevented, and they’re not counting the number of bites people received. They are only counting the number of mosquitoes were captured in a trap near a person – which might be a very poor surrogate for measuring how many cases of malaria or Zika were prevented.


Study site

They chose one of the three villages they studied in Ethiopia. Is this relevant to Zika transmission in cities in Brazil? Is it representative of the other populations in which mosquito repellents might be used?



The species of mosquito examined is one that transmits malaria in some areas, but does not transmit Zika – how certain are we that these findings would be replicated in other mosquito species?


Control group

The control they used was solvent – the same solvent that some of the tested compounds were solubilised in. Perhaps it would have been more useful to know how the test compounds compared to existing repellents. I’m not sure what a good control for a live chicken would be…


Statistical analysis

The statistical analysis was carried out post-hoc (with no mention of the analysts being blinded to the treatment groups) – the data was examined, then a method for analysing it was devised – this is known to lead to bias giving more positive findings.


Other sources of bias

Randomisation method, allocation concealment, blinding of personnel and outcome assessment, missing data and selective reporting – none were adequately described by the paper, but none seemed likely to be the source of a high risk of bias.


In summary, I think the findings of the RCT are likely to be fairly accurate. But they don’t justify the hype and the hope. This is not evidence that chickens prevent cases of Zika or malaria. When we have an RCT that looks at the mosquito species relevant to both diseases, in a variety of settings in which a repellent would be used, examining the effect of the repellent compared to the current best repellent, and examining the number of deaths, disabilities and infections, and finding at least one of the first two significantly reduced, then we can get really excited.

What’s the evidence: can cranberry products prevent UTIs?


But for more detail read this systematic review of RCTs. It’s a Cochrane systematic review, so it has a good abstract and an excellent plain language review if you don’t have much time.

cochrane cranberries

How good it the evidence?

It’s a systematic review with meta-analyses, so it’s at the top of the quality of evidence pyramid:

Evidnec pyramid

This review scores 10/11 using the AMSTAR tool to assess the quality of the systematic review. The AMSTAR tool identifies good practices that reduce the introduction of bias – all 11 areas are evidenced to be important sources of bias in systematic reviews. 10 is good. The other reviews I’ve appraised recently have scored 2-4. It gets a 10 because:

  • The review provides an a priori design
  • Duplicate study identification and data extraction were carried out
  • It undertook a comprehensive literature search
  • The review didn’t exclude unpublished data
  • A list of included and excluded studies was provided
  • Characteristics of the included studies were provided
  • Scientific quality of the included studies was assessed and documented
  • Quality of the included studies was used appropriately in formulating conclusions
  • Methods used to combine the findings of studies were appropriate
  • Conflict of interest was stated


  • The likelihood of publication bias was not assessed

Overall, we can be pretty sure that the systematic review can be trusted to give us the true answer based on the available (crappy, better than nothing, see figure 2 for their risk of bias assessment) data.

cranberries no

Not these chaps

cranberries yes

These chaps

What are the results?

The summary of results is that “cranberry products do not significantly reduce the risk of repeat symptomatic UTI compared to placebo or no treatment in groups of people at risk of repeat UTI (overall RR 0.86, 95% CI 0.71 to 1.04) or for any of the subgroups analysed.”

RR (relative risk or risk ratio) of 0.86 means that if you compared people who didn’t have cranberry with people who did have cranberry, for every 100 cases in people who didn’t get cranberry, 86 cases occurred in people who did have cranberry. However, this doesn’t mean cranberry works – take a look at the confidence intervals (CI). They span 1.00, which means that the difference seen between these groups may well have arisen by chance.

If this study was repeated, each time it was done, there would be a 95% chance that the 95% CI included the true risk. These 95% CI (0.71 to 1.04) do not show a statistically significant difference and do not exclude the possibility that cranberry products might be making things worse (104 people get cystitis instead of 100) rather than better.

The absolute risk (AR) and number needed to treat (NNT) would be much more informative than RR, but the review doesn’t provide them. Given that the reviewers used a random effects model, it would take me a bloody age to calculate them. And, the results show no difference between people taking cranberry products and those not, so calculating AR and NNT would be an academic exercise and clinically useless. And, the risk for different people in the general population varies hugely, so even if the results were statistically significant, and I could be arsed to calculate the AR and NNT, it still wouldn’t provide a useful statistic for the general population to use to understand our risk.

An aside, I think it’s worth thinking about why we aren’t more skeptical of magical fruit. Consider not buying expensive urine until there’s evidence that for you there are useful benefits (less cystitis) that outweigh harms (expending a finite financial resource, having to drink vast quantities of bitter super-staining juice every day for ever, or eat tablets every day for ever (not fun, as anyone who actually has to knows) and propping up quack shops and drug companies that prey on the trusting, vulnerable and desperate).


Should wear a mask and a striped jumper

What’s the evidence: Does Ibuprofen cause skin and blood infections in children with chickenpox?


Does Ibuprofen cause skin and blood infections in children with chickenpox?

What are doctors told?

The National Institute for Health and Clinical Evidence (NICE) Clinical Knowledge Summary “Scenario: management of an otherwise healthy child or adult with chickenpox”:1

NICE Clinical Knowledge Summary “Analgesics / Antipyretics”:2

What evidence is used to tell doctors this?Evidnec pyramid

What we’re hoping for is the highest quality of evidence to answer our question so we can be more certain in the answer. Mostly, the best evidence is at the top of the pyramid, and the quality of the evidence decreases as you move down the pyramid.

The guidelines reference three papers to support their recommendations:

Heininger and Seward, 2006

Bilj, 2010

Mikaeloff et al., 2008

Heininger and Seward, 2006 is a review, but it isn’t a systematic review, so it isn’t at the top of the evidence pyramid. We need to look at where they got their evidence from.

They reference two papers:

Lesko et al., 2001

Zerr et al., 1999

So, we can see that the guidelines for doctors are based on 4 original research papers (although these papers reference other papers too). What is the quality of this evidence?

Bilj, 2010

I couldn’t access this paper. This is very annoying. It means we can’t assess the quality of the research. It is also completely unreasonable that guidelines for treating us are based on evidence we can’t see.

Mikaeloff et al., 2008

The study design

A case-control study, half-way down the quality of evidence pyramid. Using the General Practice Research Database it looked at UK patients with chickenpox or shingles for at least two days, between 1994 and 2005. 386 patients with chickenpox had “severe skin or soft tissue complications”. They matched each of these patients with 10 of the patients who had chickenpox but no skin or soft-tissue infections. Then they worked out how much more likely the patients with skin infections were to have been given a prescription for ibuprofen.

The results

12 of 386 patients with severe skin or soft tissue complications took ibuprofen. 14 of 2402 patients without severe skin or soft tissue complications took ibuprofen. The relative risk is 5.2 (patients taking ibuprofen were 5x more likely to have severe skin or soft tissue complications). The absolute risk increase is 0.025 (patients taking ibuprofen increased their risk of severe skin or soft tissue complications by 0.025). The number of patients that would have to be prescribed ibuprofen for one patient to be harmed is 40.

The problems

Case-controls studies are subject to more and greater biases than research further up the pyramid. The patients in this study are adults and children – they have an average age of about 11 years old, but we don’t know the age of the patients who developed skin or soft-tissue infections – and we want to know the answer for children. They found 386 patients with severe skin or soft tissue complications, of whom only 26 had taken ibuprofen. This is very few patients if you want to work out the role of ibuprofen, which, if it is a risk factor, is likely to be a very small risk factor. With numbers this small the results are highly subject to chance. The way that they found out who took ibuprofen was to look at who had been prescribed ibuprofen – this will over-count patients who received a prescription but didn’t collect it from the pharmacy or didn’t take it, and will not count patients who buy their ibuprofen over the counter (30p a packet).

Can we trust the results?

I think not. This study ignored patients who took ibuprofen bought over the counter, which is how almost all of us access ibuprofen. Patients who were more ill, and more likely to develop skin or soft-tissue infections would have been more likely to see a GP and to have a record of a prescription for ibuprofen. These factors will make ibuprofen look much worse than it is. The number of patients is very very small (26 who were prescribed ibuprofen). It is highly likely that this result has arisen by chance.


Lesko et al., 2001

The study design

The study design is a case-control study, half-way down the quality of evidence pyramid. They included children who had been admitted to hospital in the USA, with chickenpox and “invasive or necrotising soft tissue infection”, between 1996 and 1998. 52 patients with chickenpox had invasive Group A streptococcal infection. They matched each of these patients with 4 of the patients who had chickenpox but invasive Group A streptococcal infection. Then they worked out how much more likely the patients with skin infections were to have been given a prescription for ibuprofen.

The results

18 of 52 patients with invasive Group A streptococcal infection took ibuprofen. 36 of 172 patients without invasive Group A streptococcal infection took ibuprofen. The relative risk is 1.7 (patients taking ibuprofen were nearly 2x more likely to have invasive Group A streptococcal infection). The absolute risk increase is 0.14 (patients taking ibuprofen increased their risk of severe skin or soft tissue complications by 0.14). The number of patients that would have to be prescribed ibuprofen for one patient to be harmed is 7.

The problems

Case-controls studies are subject to more and greater biases than research further up the pyramid. The cases in this study have been hospitalised but control patients haven’t. It seems likely that sicker children are more likely to be hospitalised, and more likely to have been given more medication (including ibuprofen). They found 52 patients with severe skin or soft tissue complications, of whom only 18 had taken ibuprofen. This is very few patients if you want to work out the role of ibuprofen, which, if it is a risk factor, is likely to be a very small risk factor. With numbers this small the results are highly subject to chance. The way that they found out who took ibuprofen was either to examine medical records (cases) or to interview the parents (controls). Asking people what they remember giving to their child before something dramatic (hospitalisation) didn’t happen, is not likely to be a reliable source of data, and exposure (to ibuprofen) may well be underreported.

Can we use and trust the results?

Maybe. If sicker children are more likely to be hospitalised and more likely to have been given more medication then these results will make ibuprofen look worse than it is. The number of patients is very small (54 who took or reported taking ibuprofen). It is quite possible that this result has arisen by chance.


Zerr et al., 1999

The study design

A case-control study, half-way down the quality of evidence pyramid. It compares children hospitalised for necrotising fasciitis who recently had chickenpox, with children hospitalised for a “different soft tissue infection” who recently had chickenpox. Between 1993 and 1994 48 patients were examined. Then they worked out how much more likely the patients with necrotising faciitis were to have taken ibuprofen.

The results

9 of 19 patients with necrotising fasciitis took ibuprofen. 4 of 26 patients with a different soft tissue infection took ibuprofen. The relative risk is 3.2 (patients taking ibuprofen were 3x more likely to have necrotising faciitis than a different soft tissue infection). The absolute risk increase is 0.32 (patients taking ibuprofen increased their risk of necrotising fasciitis by 0.32). The number of patients that would have to be prescribed ibuprofen for one patient to get necrotising fasciitis is 3.

The problems

Case-controls studies are subject to more and greater biases than research further up the pyramid. All these patients had an infection – so we are comparing different types of infections – but what we want to know is whether ibuprofen causes any infection. They found 48 patients with skin or soft tissue complications, of whom only 13 had taken ibuprofen. This is very very few patients if you want to work out the role of ibuprofen, which, if it is a risk factor, is likely to be a very small risk factor. And, as we saw before, this study isn’t even asking the question we want to answer – which is whether ibuprofen increases any infection. With numbers this small the results are highly subject to chance. This study tests 20 different risk factors. The more risk factors you look at, the greater the chance of one of them giving a positive result by chance. P=0.05 is commonly used as a measure of significance. Using this p value there is a 1/20 chance that a positive result will arrive by chance. This paper has looked at 20 outcomes… so we would expect that one would be positive just by chance.

Can we use and trust the results?

I think not. The question being tested is related to, but is not the question we want to answer. The number of patients is very very small (13 who were prescribed ibuprofen), and the authors examined 20 risk factors. It is highly likely that this result has arisen by chance.

In summary

I don’t think that any of these three studies is reliable. The hidden study might be better, but how do we know? All three visible studies give a hint that ibuprofen might be a problem. But none of them show that ibuprofen is a risk-factor for skin and blood infections in children with chickenpox, and they don’t demonstrate that ibuprofen causes skin and blood infections. The clinical guidelines (which doctors often rely upon) are based on very unreliable evidence. There is other evidence out there but it will take longer than I have to trawl through. Bearing in mind that NICE are probably better equipped for this than me, and that these are the papers they referenced, I think it is unlikely that any better quality research exists.


So, what’s the evidence? Does Ibuprofen cause skin and blood infections in children with chickenpox?

We don’t know. There is very shaky clinical evidence to suspect it might, and no reliable clinical evidence to show it does.

STILL NOT A DOCTOR. STILL NOT MEDICAL ADVICE. Hope it makes you think though.

Grosse Point Blank

While visiting Eccles with my brother, I found this plaque on a pub.


It’s a reminder of one of the outbreaks of cholera that occurred in the UK, many of which were epidemics killing hundreds or thousands of people. The pub isn’t lovely, but that’s the current-Eccles-factor rather than the past-cholera-factor.

The story of cholera in the UK is special, because it is also the story of the birth of epidemiology, and a part of the story of the development of our understanding of disease transmission.

In 1854 there was an outbreak of cholera in Soho, in London. As a skeptic of the miasma theory (that bad air caused disease), John Snow (a doctor from York, handy with anaesthetics) proposed an alternative theory; that it might be transmitted by polluted water. His skepticism and theory were shared by others working contemporaneously with him, and before him, although he was not aware of much of their work.

Not this Jon Snow.


Not this Jon Snow.


This John Snow.

John SNow


John Snow wrote an essay on this theory of cholera outbreaks in 1849, helped found the Epidemiological Society of London in 1850, and by observing and graphing the pattern of the outbreak in 1854 was able to prove the source of the outbreak. He used a new(ish) type of graph (used years before by Thomas Shapter) – a dot map – with a dot in the geographical location of each case of cholera. By observing the distribution of the dots (cases) and by interviewing residents he was able to find a pattern in the location and behaviour of people who had contracted cholera – they all used a public water pump on Broad Street.


The outbreak was subsiding by the time John Snow had enough evidence to act (people tend not to stick around if there’s a reasonably high chance of a horrible death). He had the handle of the pump removed, and the cases of cholera stopped. Later it was discovered that a cesspit was leaking into the public well. Drinking faeces was not a fashionable hobby then (unlike today), so rather than admit that was what had been happening, the pump handle was replaced so that everyone could quietly keep doing it.

And we’re still doing it! In the last two months there have been cholera outbreaks in Mozambique (Mocuba), Iraq (Baghdad), Muscat (Iran), Uvinza, Mwanza and Dar es Salaam (Tanzania), Kye-thi (Myanmar), Naivasha (Kenya) and South Kivu (Democratic Republic of the Congo). Those in Iraq and Tanzania have been particularly bad, killing many thousands of people.

Outbreaks of cholera can mostly be prevented by separating sewage (dirty) from the drinking water supply (ideally clean), but clean water and toilets are not exciting. Money invested in development projects and given to charities providing these facilities goes a long way. The problem is we’re not into evidence-based giving. And we like donkeys better than people anyway. And cholera is not glam. It’s been around for a very long time. It’s not exciting looking. It’s totally preventable. It’s totally treatable. BORING. And you basically vomit and poo yourself to death. GROSS.

Zika anyone? zika

We’d rather the antimicrobial apocalypse than be denied useless antibiotics


This research came out this week…

Antibiotic prescribing and patient satisfaction in
primary care in England:
cross-sectional analysis of national patient survey data and prescribing data

…showing that antibiotic prescribing volume is a significant positive predictor of ‘doctor satisfaction’ and ‘practice satisfaction’…


…antibiotic prescribing volume is the single strongest positive predictor (of 13 prescribing variables) of overall satisfaction.

Table 3 has some other interesting data on prescribing practise and patient satisfaction.

This is an observational study, and can’t prove a relationship, but it is a good clue. More research required, as (almost) ever…