日本版敗血症診療ガイドライン2020 (J-SSCG2020) の血糖管理でネットワークメタアナリシスをやってみる

The Japanese Clinical Practice Guidelines for Management of Sepsis and Septic Shock 2020 (J-SSCG 2020)
「日本版敗血症診療ガイドライン2020 (J-SSCG2020)」正式版 公開のお知らせ|日本集中治療医学会
日本版敗血症診療ガイドライン2020 (J-SSCG2020)」正式版 公開のお知らせ
集中治療学会のほうは「許可なくリンクを貼ることを禁じます」と書いてあるが、ネットに無償公開しているものについて引用の範囲でリンクを貼ることについては法律的になんら問題はないので、APRINで倫理講習を受けましょう、とかアホなこと言ってないで法律的なことを真面目に勉強したほうがいいと思う。

ネットワークメタアナリシスというものがあり、ガイドライン中のいくつかのCQで検討が行われている。
CQ13-2 の「敗血症患者の目標血糖値はいくつにするか?」について、35件の研究が引用されている。
このうち、23番目のKalfon 2014 の論文では、付録の III >180 vs <110 アウトカム①短期死亡について、リスク比が Not estimated となっている。これは、<110 群で1336人にすべきところを136人と誤って解析しているからであるが、これを正しく1336人にして解析すると、論文と同じようなそれっぽい解析結果は出るが、対照 <110 群に対して144-180 群の結果が0.90 [0.83 - 0.97] となっているが、正しい入力データだと0.8946であり有効数字を考慮すると0.90 にならない。

Treatment estimate (sm = 'RR', comparison: other treatments vs '<110'):
            RR           95%-CI     z p-value
<110         .                .     .       .
>180    1.0102 [0.9480; 1.0766]  0.31  0.7538
110-144 0.8841 [0.7144; 1.0942] -1.13  0.2575
144-180 0.8946 [0.8272; 0.9676] -2.78  0.0054

Kalfon の誤った入力データでやってみるともちろん出力は異なるので、本解析と付録で入力データが異なるし、出力も完全再現に至らないので本当どういうことなんだろう()

出力結果のグラフの点の大きさやエッジの太さを適当にいじったが、本文に書いていないので本当にあっているかは謎。
Control size of points in netgraph in R - Stack Overflow

library(NMA)
library(netmeta)

dat <- read.csv(text="
study,trial,trt,n,d,pubyear
1,Oksanen,110-144,51,18,2007
1,Oksanen,<110,39,13,2007
2,Finfer,144-180,3012,751,2009
2,Finfer,<110,3010,829,2009
3,Preiser,144-180,542,105,2009
3,Preiser,<110,536,125,2009
4,Green,144-180,36,9,2010
4,Green,<110,45,16,2010
5,Coester,144-180,40,4,2010
5,Coester,<110,39,2,2010
6,Cappi,144-180,35,10,2012
6,Cappi,<110,28,5,2012
7,van den Berghe,>180,765,55,2001
7,van den Berghe,<110,783,85,2001
8,Mitchell,>180,35,3,2006
8,Mitchell,<110,35,9,2006
9,van den Berghe,>180,605,242,2006
9,van den Berghe,<110,595,222,2006
10,Iapichino,>180,36,11,2008
10,Iapichino,<110,36,13,2008
11,De La Rosa Gdel,>180,250,96,2008
11,De La Rosa Gdel,<110,254,102,2008
12,Brunkhorst,>180,288,102,2008
12,Brunkhorst,<110,247,98,2008
13,Arabi,>180,257,83,2008
13,Arabi,<110,266,72,2008
14,Savioli,>180,45,13,2009
14,Savioli,<110,45,14,2009
15,Annane,>180,254,109,2010
15,Annane,<110,255,117,2010
16,Arabi,>180,120,45,2011
16,Arabi,<110,120,42,2011
17,Kalfon,>180,1312,447,2014
17,Kalfon,<110,1336,431,2014
18,Wang,>180,44,14,2017
18,Wang,<110,44,12,2017
19,McMullin,144-180,9,4,2007
19,McMullin,110-144,11,6,2007
20,Davies,>180,34,6,1991
20,Davies,110-144,35,6,1991
21,Walters,>180,12,0,2006
21,Walters,110-144,13,1,2006
22,Farah,>180,48,26,2007
22,Farah,110-144,41,19,2007
23,Bruno,>180,15,0,2008
23,Bruno,110-144,31,2,2008
24,Bilotta,>180,49,4,2008
24,Bilotta,110-144,48,4,2008
25,Chan,>180,55,3,2009
25,Chan,110-144,54,2,2009
26,de Azevedo,>180,169,42,2010
26,de Azevedo,110-144,168,38,2010
27,Hsu,>180,57,28,2012
27,Hsu,110-144,55,18,2012
28,Giakoumidakis,>180,107,7,2013
28,Giakoumidakis,144-180,105,1,2013", stringsAsFactors=FALSE)

hf <- setup(study=study, trt=trt, d=d, n=n, measure="RD", ref="<110", data=dat)
n <- nma(hf, eform=TRUE)

nmadat <- as.data.frame(t(sapply(split(dat[,-1], dat$study), unlist))[,c(-1, -10)])
for(i in 4:7){
  nmadat[,i] <- as.numeric(nmadat[,i])
}

p1 <- pairwise(treat=list(trt1, trt2), event=list(d1, d2), n=list(n1, n2), dat=nmadat, sm="RR")
net1 <- netmeta(p1, common = FALSE)
net1
cols <- c("skyblue3", "orange", "red", "green")
netgraph(net1, points=TRUE, cex=1.5, number.of.studies=TRUE, thickness="se.random", col.points=cols, cex.points=table(c(net1$treat1, net1$treat2)), plastic=FALSE, rotate=-90)

1. Corticosteroid treatment and intensive insulin therapy for septic shock in adults: a randomized controlled trial
2. Intensive versus conventional insulin therapy: a randomized controlled trial in medical and surgical critically ill patients
3. Permissive underfeeding and intensive insulin therapy in critically ill patients: a randomized controlled trial
4. The effect of intensive insulin therapy on infection rate, vasospasm, neurologic outcome, and mortality in neurointensive care unit after intracranial aneurysm clipping in patients with acute subarachnoid hemorrhage: a randomized prospective pilot trial
5. Intensive insulin therapy after severe traumatic brain injury: a randomized clinical trial
6. Safety and efficacy of intensive insulin therapy in critical neurosurgical patients
7. Intensive versus modified conventional control of blood glucose level in medical intensive care patients: a pilot study
8. Intensive insulin therapy and pentastarch resuscitation in severe sepsis
9. Treatment of hyperglycemia in ischemic stroke (THIS): a randomized pilot trial
10. Dyslipidemia: a prospective controlled randomized trial of intensive glycemic control in sepsis
11. Intensive perioperative glucose control does not improve outcomes of patients submitted to open-heart surgery: a randomized controlled trial
12. Intensive Insulin Therapy in Severe Traumatic Brain Injury:... : Journal of Trauma and Acute Care Surgery
13. Metabolic control in diabetic subjects following myocardial infarction: difficulties in improving blood glucose levels by intravenous insulin infusion
14. A carbohydrate-restrictive strategy is safer and as efficient as intensive insulin therapy in critically ill patients
15. Strict glycaemic control in patients hospitalised in a mixed medical and surgical intensive care unit: a randomised clinical trial | Critical Care | Full Text
16. Insulin therapy of hyperglycemia in intensive care
17. Intensive versus conventional glucose control in critically ill patients
18. Effects of intensive glycemic control on outcomes of cardiac surgery
19. Intensive versus conventional insulin therapy in critically ill neurologic patients
20. Reduction of nosocomial infections in the surgical intensive-care unit by strict glycemic control
21. Moderate glucose control results in less negative nitrogen balances in medical intensive care unit patients: a randomized, controlled study
22. Tight glycemic control does not affect asymmetric-dimethylarginine in septic patients
23. Tight computerized versus conventional glucose control in the ICU: a randomized controlled trial
24. Lowering of glucose in critical care: a randomized pilot trial
25 A phase II randomised controlled trial of intensive insulin therapy in general intensive care patients
26. Strict versus moderate glucose control after resuscitation from ventricular fibrillation
27. Cerebral metabolic effects of strict versus conventional glycaemic targets following severe traumatic brain injury
28. A prospective randomised multi-centre controlled trial on tight glucose control by intensive insulin therapy in adult intensive care units: the Glucontrol study
29. Tight glycemic control may favor fibrinolysis in patients with sepsis
30. Intensive insulin treatment reduces transient ischaemic episodes during acute coronary events in diabetic patients
31. Randomized Controlled Trial of Intensive Versus Conservative Glucose Control in Patients Undergoing Coronary Artery Bypass Graft Surgery: GLUCO-CABG Trial
32. Intensive insulin therapy in critically ill patients
33. Intensive insulin therapy in the medical ICU
34. A randomised, controlled pilot study to investigate the potential benefit of intervention with insulin in hyperglycaemic acute ischaemic stroke patients
35. Intensive insulin therapy for preventing postoperative infec... : Medicine