Okumura Y, Nishi D: Risk of recurrent overdose associated with prescribing patterns of psychotropic medications after nonfatal overdose. Neuropsychiatric Disease and Treatment 13: 653-665, 2017.
https://doi.org/10.2147/NDT.S128278
13. ⽣存時間分析の特徴
関連ガイドライン
生存時間分析の適正報告調査
Abraira V et al: Reporting quality of survival analyses in medical
journals still needs improvement. A minimal requirements proposal (J
Clin Epidemiol, in press)
Altman DG et al: Review of survival analyses published in cancer
journals (Br J Cancer 72: 511-8, 1995)
生存時間分析のReporting Guideline
Allison PD: Survival analysis (The reviewer’s guide to quantitative
methods in the social sciences: 413-424, 2010)
Lang TA, Secic M: Assessing time-to-Event as an endpoint: reporting
survival analysis (How to report statistics in medicine: 115-124, 2006)
予後研究のReporting Guideline
Hayden JA et al: Ann Intern Med. 2013 Feb 19;158(4):280-6.
13
14. ⽣存時間分析の特徴
報告すべき点
報告すべき点
文献
報告すべき点
文献
1. 研究集団の定義
2. 追跡期間の定義
3. イベントの定義
4. 打ち切りの情報
5. 共変量の説明
6. 時間依存共変量
[2]
[1,3]
[1-3]
[1,3]
[1-2]
[1]
10. 欠測データの記述
11. 要約統計量
12. 図示
13. 比例ハザード性の検討
14. ハザード比と信頼区間
15. モデル比較
[1-3]
[1,3]
[1]
[1,3]
[1,3]
[3]
7. 競合リスク
8. 例数設計
9. ソフトウェア
[1]
[1,3]
[1]
16. 予測モデルの精度
[3]
[1] Allison PD: Survival analysis (The reviewer’s guide to quantitative methods in the social sciences: 413-424, 2010)
[2] Hayden JA et al: Ann Intern Med. 2013 Feb 19;158(4):280-6.
[3] Abraira V et al: J Clin Epidemiol, in press.
14
17. 共通項⽬ > 研究集団の定義/追跡期間の定義/イベントの定義/打ち切りの情報
記載事例
方法 (母集団の特性)
» The cohort comprised patients admitted to ICUs in the United States
included in the Project IMPACT database (Cerner Corporation, Kansas
City, Missouri), which prospectively collects information on patients in
194 ICUs at 131 hospitals. Project IMPACT is a voluntary, fee-based,
clinical information system comprising a large, diverse sample of ICUs
that is commonly used for critical care outcomes research (16–19).
Wagner J et al: Ann Intern Med. 2013 Oct 1;159(7):447-55.
17
18. 共通項⽬ > 研究集団の定義/追跡期間の定義/イベントの定義/打ち切りの情報
記載事例
方法 (抽出法と症例登録法)
» The sampling frame for the NIS [Nationwide Inpatient Sample]
composes approximately 90% of all hospital discharges in the United
States. Sampling from the universe of US hospitals is stratified based
on 5 hospital characteristics: US region, urban or rural location,
ownership or control, bed size, and teaching status. [1]
» Data collectors who are certified by Project IMPACT capture detailed
clinical information at each site from the time of ICU admission until
hospital discharge.[2]
[1] Okumura MJ et al: Arch Pediatr Adolesc Med. 2006 Oct;160(10):1054-60.
[2] Wagner J et al: Ann Intern Med. 2013 Oct 1;159(7):447-55.
18
19. 共通項⽬ > 研究集団の定義/追跡期間の定義/イベントの定義/打ち切りの情報
記載事例
方法 (登録期間)
» We defined a cohort of patients initiating oral mono-therapy for diabetes
between 1 October 2001 and 30 September 2008 using data sets from
national Veterans Health Administration (VHA) Decision-Support
Services: (後略)
Roumie CL et al: Ann Intern Med. 2012 Nov 6;157(9):601-10.
19
20. 共通項⽬ > 研究集団の定義/追跡期間の定義/イベントの定義/打ち切りの情報
記載事例
方法 (登録場所)
» After institutional research ethics approval, preoperative, intraoperative,
and postoperative data on individuals undergoing cardiac surgery at the
Toronto General Hospital (Toronto, Ontario) and University of Ottawa
Heart Institute (Ottawa, Ontario) were prospectively collected in distinct
hospital-specific registries.
Wijeysundera DN et al: JAMA. 2007 Apr 25;297(16):1801-9.
20
21. 共通項⽬ > 研究集団の定義/追跡期間の定義/イベントの定義/打ち切りの情報
記載事例
方法 (適格基準)
» The study population comprised veterans aged 18 years or older who
received regular VHA care (a VHA encounter or prescription fill at least
once every 180 days) for at least the past 365 days. Incident users with
known birth date and sex and with more than 365 days of baseline data
preceding their first eligible prescription fill were identified. Patients
were eligible if they filled a first prescription for an oral antidiabetic drug
after at least 365 days without any oral or injectable diabetic drug fill
(new users) (12).
Roumie CL et al: Ann Intern Med. 2012 Nov 6;157(9):601-10.
21
22. 共通項⽬ > 研究集団の定義/追跡期間の定義/イベントの定義/打ち切りの情報
記載事例
方法 (参加割合)
» Of the 6299 cases identified, 5042 provided written informed consent
and participated in the study (participation rate, 80.0%). Among the
remaining cases, 757 (12.0%) refused to participate, 258 (4.1%) were
absent during study enrollment, 83 (1.3%) could not be contacted, and
159 (2.5%) were excluded for other miscellaneous reasons such as
health or communication problems.
Shu XO et al: JAMA. 2009 Dec 9;302(22):2437-43.
22
27. 共通項⽬ > 研究集団の定義/追跡期間の定義/イベントの定義/打ち切りの情報
ゼロ時点 ≠ 追跡開始時点の事例集
ゼロ時点
上皮性卵巣癌の診断時
生誕
最終月経開始日
C型肝炎の1度目の肝
生検時
追跡開始時点
研究参加時
研究参加時
妊娠の診断
イベント発生時点
診断後5年以内の死亡
認知症の発症
自然流産
治療1年以上経過 肝臓関連の合併症
後の肝生検時
[1] Bolton KL et al: JAMA. 2012 Jan 25;307(4):382-90.
[2] van Himbergen TM et al: Arch Neurol. 2012 May;69(5):594-600.
[3] Chambers CD et al: Vaccine. 2013 Oct 17;31(44):5026-32.
[4] Mallet V et al: Ann Intern Med. 2008 Sep 16;149(6):399-403.
文献
[1]
[2]
[3]
[4]
27
28. 共通項⽬ > 研究集団の定義/追跡期間の定義/イベントの定義/打ち切りの情報
記載事例
方法 (追跡開始と終了の時点/ゼロ時点)
» This was a retrospective cohort study of individuals recorded in the
United States Renal Data System (USRDS) database who initiated
ESKD[End-Stage Kidney Disease] treatment with dialysis for the first
time at younger than 21 years of age between January 1, 1990, and
December 31, 2010, and who were followed up until December 31,
2010.
» Time zero was the day of dialysis initiation, and observation was
censored at first transplant, death, or end of observation.
Mitsnefes MM et al: JAMA. 2013 May 8;309(18):1921-9
28
29. 共通項⽬ > 研究集団の定義/追跡期間の定義/イベントの定義/打ち切りの情報
記載事例
結果 (打ち切りデータの追跡期間の要約)
» Follow-up to death or at least year 3 is now 98% complete (3062/3120),
and median follow-up in survivors is 9 years (IQR 6–11).
Halliday A et al: Lancet. 2010 Sep 25;376(9746):1074-84.
29
31. 共通項⽬ > 研究集団の定義/追跡期間の定義/イベントの定義/打ち切りの情報
記載事例
方法 (操作的定義,信頼性と妥当性)
» The primary composite outcome was hospitalization for acute
myocardial infarction (AMI) or stroke, or death. We defined “AMI” as an
ICD-9-CM primary discharge diagnosis for fatal and nonfatal AMI (ICD9-CM code 410.x) (positive predictive value, 67% to 97% compared
with chart review) (13–15). ...(中略)... We determined mortality using the
VHA Vital Status File, which combines information from multiple
sources (Medicare, the VHA, the U.S. Social Security Administration,
and VHA compensation and pension benefits) to determine date of
death (sensitivity, 98.3%; specificity, 99.8%; relative to the National
Death Index) (17).
Roumie CL et al: Ann Intern Med. 2012 Nov 6;157(9):601-10.
31
32. 共通項⽬ > 研究集団の定義/追跡期間の定義/イベントの定義/打ち切りの情報
記載事例
方法 (操作的定義,測定法の一貫性)
» The outcome for this study was time to first CHD event, a composite of
nonfatal MI and fatal CHD. The Appendix provides details on the very
similar standardized definitions of CHD events for the 2 cohorts (27,
28).
Cardiovascular Health
Studyの定義
Koller MT et al: Ann Intern Med. 2012 Nov 6;157(9):601-10.
Rotterdam Studyの定義
32
33. 共通項⽬ > 研究集団の定義/追跡期間の定義/イベントの定義/打ち切りの情報
記載事例
方法 (操作的定義,イベントの反復)
» We first evaluated LOS[length of stay] (in hours) during the initial ICU
admission. Length of stay was calculated as the number of hours from
ICU admission to ICU discharge. ...(後略)...
» To promote data independence, we analyzed only initial ICU discharges
within a hospitalization. However, because Project IMPACT does not
track patients across hospitalizations, those who were discharged from
the hospital and then readmitted may have contributed more than 1 ICU
admission.
Wagner J et al: Ann Intern Med. 2013 Oct 1;159(7):447-55.
33
38. 共通項⽬ > 研究集団の定義/追跡期間の定義/イベントの定義/打ち切りの情報
記載事例
方法 (種類/理由/量)
» Patients were censored if they switched study medications,
discontinued treatment (in which case they were followed for 100 days
from their last prescription to identify events that may have precipitated
discontinuation), commenced treatment with a different thiazide, or
reached the end of the study (31 March 2010). We followed patients for
a maximum of 5 years.
Dhalla IA et al: Ann Intern Med. 2013 Mar 19;158(6):447-55.
38
39. 共通項⽬ > 研究集団の定義/追跡期間の定義/イベントの定義/打ち切りの情報
記載事例
方法 (種類/理由/量)
» We followed all individuals from baseline until the occurrence of a
venous thromboembolic event or until censoring. Baseline was defined
as the date at which the participant was first selected for inclusion into
the Copenhagen City Heart Study. ...(中略)... We terminated any further
follow-up on 31 December 1999 because this was the last date on
which complete diagnostic information on end points was obtained.
» Seventy-six participants emigrated during follow-up and were therefore
censored at the emigration date. Four participants who could not be
traced were censored at the date on which they were lost.
Consequently, follow-up is more than 99% complete. Median follow-up
time was 23 years (range, 0.04 to 23 years).
Juul K et al: Ann Intern Med. 2004 Mar 2;140(5):330-7
39
52. 時間依存共変量 > 独⽴変数の影響の経時変化の検討/時間依存バイアスへの対処
記載事例
方法 (比例ハザード性/時間依存共変量)
» The proportional hazards assumption for the model was checked by
examining log minus-log transformed Kaplan-Meier estimates of the
survival functions for the atypical and typical antipsychotic users groups
plotted against time to death/follow-up time. These curves help in
identifying non-proportionality patterns in hazard function such as
convergent (difference in risk between the 2 groups decreases with
time), divergent, or crossing of the curves. In addition, Schoenfeld test
for the violation of proportional hazards, which assess the correlation
between scaled residuals and time, was also conducted. As the
proportional-hazard assumptions were violated in the above-mentioned
diagnostic test, the extended Cox hazard model was used for timevarying exposure of the treatment variable.
Aparasu RR et al: Med Care. 2012 Nov;50(11):961-9.
52
53. 時間依存共変量 > 独⽴変数の影響の経時変化の検討/時間依存バイアスへの対処
記載事例
方法 (時間依存共変量)
» The Heaviside function was used that yields constant hazard ratios
(HR) for different time intervals.38 According to the graphical depiction of
the log minus-log transformed Kaplan-Meier curve and previous
literature,12,16 the follow-up period was divided into 2 time intervals (<
40 d, 40–180d) after the initiation of antipsychotic treatment and 2
different HRs were obtained.
Aparasu RR et al: Med Care. 2012 Nov;50(11):961-9.
53
54. 時間依存共変量 > 独⽴変数の影響の経時変化の検討/時間依存バイアスへの対処
記載事例
結果 (ハザード比と信頼区間)
» Table 2 presents results from the Cox proportional-hazard model and
the extended Cox proportional-hazard model using the Heaviside
function. A statistically significant difference was found [HR, 1.41; 95%
confidence interval (CI), 1.27–1.57] between atypical and typical
antipsychotic users with respect to risk of death. This can be considered
as the average risk of mortality within 180 days of exposure to typical
agents.
Aparasu RR et al: Med Care. 2012 Nov;50(11):961-9.
54
55. 時間依存共変量 > 独⽴変数の影響の経時変化の検討/時間依存バイアスへの対処
記載事例
結果 (ハザード比と信頼区間)
» As there was the violation of the proportional hazards assumption, the
extended Cox proportional-hazard model was used. The analysis
revealed that risk of death was 80% more in nursing home residents
using typical antipsychotics within 40 days of antipsychotic treatment
(HR, 1.81; 95% CI, 1.49–2.18) than those using atypical agents. A
moderate increase in risk (HR, 1.24; 95% CI, 1.09–1.42) was observed
from 40 to 180 days of exposure to typical antipsychotic agents.
Aparasu RR et al: Med Care. 2012 Nov;50(11):961-9.
55
60. 時間依存共変量 > 独⽴変数の影響の経時変化の検討/時間依存バイアスへの対処
時間依存バイアス狩り
食道がん
» We read with interest the study published by Kastelein et al1 in a recent
issue of GASTROENTEROLOGY. The authors investigated the effects
of nonsteroidal anti-inflammatory drugs (NSAIDs) and statins on the risk
of neoplastic progression from Barrett’s esophagus (BE). ...(中略)...
These impressive risk reductions are likely owing to immortal persontime bias.2
» Thus, before concluding that NSAIDs and statins reduce the risk of
neoplastic progression in patients with BE, it would be informative if the
authors redo their analyses by employing the appropriate techniques to
define exposure in a time-dependent fashion.
Azoulay L, Suissa S: Gastroenterology. 2012 May;142(5):e20-1
60
61. 時間依存共変量 > 独⽴変数の影響の経時変化の検討/時間依存バイアスへの対処
時間依存バイアス狩り
冠動脈バイパス術
» We read with great interest the article by Kim and associates1 in a
recent issue of the Journal of Thoracic and Cardiovascular Surgery. We
find this topic area engaging and relevant to clinical practice, especially
in light of expanding indications of dual antiplatelet therapy and recent
advances in perioperative management lending to improved acute
operative mortality.2 However, we have important concerns regarding
the methodology and statistical analyses undertaken.
» It would be informative for the authors to treat clopidogrel exposure in a
time-dependent fashion to more accurately dissect the role of dual
antiplatelet therapy after CABG.
Vaduganathan M, Suissa S: J Thorac Cardiovasc Surg. 2011 Mar;141(3):850-1;
61
62. 時間依存共変量 > 独⽴変数の影響の経時変化の検討/時間依存バイアスへの対処
時間依存バイアス狩り
多発性硬化症
» We read with interest Trojano and colleagues’1 observational study
reporting an important decrease in the rates of secondary progression,
Expanded Disability Status Scale scores of 4 and 6 for patients treated
with interferon (IFN) compared with nontreated patients (adjusted rate
ratios: 0.38, 0.70, and 0.60, respectively).
» We believe that the study is subject to immortal time bias, which has
been described in other study cohorts of drug effects in several
diseases.2–4
Renoux C, Suissa S: Ann Neurol. 2008 Jul;64(1):109-10
62
63. 時間依存共変量 > 独⽴変数の影響の経時変化の検討/時間依存バイアスへの対処
記載事例
方法 (時間依存共変量)
» Cox proportional hazards regression was used to test the hypothesis
that NRT[nicotine replacement therapy] use was associated with a
change in ICU or hospital mortality risk over time. Potential confounders
included age, APACHE II score, gender, and history of significant
alcohol use. The co-primary endpoints were ICU and hospital mortality.
Because NRT could be started after ICU admission, with potential for
self-selecting survivors (so-called immortal time bias), NRT was entered
as a discrete time-varying covariate[12].
Gillies MA et al: Intensive Care Med. 2012 Oct;38(10):1683-8.
63
64. 時間依存共変量 > 独⽴変数の影響の経時変化の検討/時間依存バイアスへの対処
記載事例
結果 (ハザード比)
» The Cox models (Table 2; Fig. 1) showed that, after adjustment for
APACHE risk, age, sex and alcohol use, the hazard ratios for NRT
administration were not statistically significant for both ICU and hospital
mortality.
Gillies MA et al: Intensive Care Med. 2012 Oct;38(10):1683-8.
64
66. 競合リスク > 特徴/独⽴変数の影響の検討/発⽣確率の検討
競合リスク
記載事項 (抜粋)
競合リスクのモデルの説明 [1]
ハザード比と信頼区間[1,2]
図示[1]
[1] Allison PD: Survival analysis (The reviewer’s guide to quantitative methods in the social sciences: 413-424, 2010)
[2] Abraira V et al: J Clin Epidemiol, in press.
66
67. 競合リスク > 特徴/独⽴変数の影響の検討/発⽣確率の検討
競合リスクの事例集
調査対象
腹膜透析患者
主イベント
競合リスク
初回の腹膜炎発症 血液透析に変更/腎移植/
死亡
肝細胞癌患者 がんの再発
再発前の死亡
薬剤変更/多剤併用療法
抗うつ薬の単剤 治療中断
治療開始患者
統合失調症の 統合失調症の発症 前駆症状の消失
ハイリスク群
追跡の脱落
AIDS発症/死亡
治療開始した
HIV感染者
[1] Teixeira L et al: BMC Nephrol. 2013 May 24;14(1):110.
[2] Wu CY et al: JAMA. 2012 Nov 14;308(18):1906-14.
[3] Wu CS et al: Gen Hosp Psychiatry. 2013 May-Jun;35(3):279-85.
[4] Schlosser DA et al: Schizophr Bull. 2012 Nov;38(6):1225-33.
[5] Shepherd BE et al: Am J Epidemiol. 2013 Sep 1;178(5):819-28.
文献
[1]
[2]
[3]
[4]
[5]
67
74. 競合リスク > 特徴/独⽴変数の影響の検討/発⽣確率の検討
記載事例
方法 (競合リスクのモデル)
» To examine differential associations of aspirin use with colorectal
cancer risk by tumor molecular subtype, we used Cox proportional
cause-specific hazards regression model with a duplication method for
competing risks data. This method permits estimation of separate
associations of a risk factor (ie, aspirin use) with each tumor subtype,
and has been used to assess whether a risk factor has statistically
different regression coefficients for different tumor subtypes.5,25
Nishihara R et al: JAMA. 2013 Jun 26;309(24):2563-71.
74
77. 競合リスク > 特徴/独⽴変数の影響の検討/発⽣確率の検討
記載事例
方法 (競合リスクのモデル)
» Other-cause mortality was modeled with the proportional subdistribution
hazards regression described by Fine and Gray (15), with prostate
cancer death treated as a competing risk. A competing event eliminates
the possibility of the primary event of interest, and treating it as a
censored observation would violate the noninformative censoring
assumption of the Cox proportional hazards model.
Daskivich TJ et al: Ann Intern Med. 2013 May 21;158(10):709-17.
77
78. 競合リスク > 特徴/独⽴変数の影響の検討/発⽣確率の検討
記載事例
結果 (図示)
» Among these[3 or more comorbidity condition], men younger than 60
years, aged 61 to 74 years, and older than 75 years at diagnosis had
10-year other-cause mortality rates of 26% (17 of 65), 40% (108 of
272), and 71% (58 of 82), respectively.
その他の死亡は
併存症が多いと増える
Daskivich TJ et al: Ann Intern Med. 2013 May 21;158(10):709-17.
前立腺がん死亡は少ない
78
80. 推薦資料
推薦図書
はじめての競合リスク
Competing risks: A practical perspective
600ページを超えた第3版
Survival Analysis: A Self-Learning Text
鉄板
Applied Survival Analysis: Regression
Modeling of Time to Event Data
80
81. 推薦資料
推薦論文
図示の美の追求
Pocock SJ et al: Lancet.
2002;359(9318):1686-9.
時間依存バイアス入門
Lévesque LE et al: BMJ. 2010;340:b5087.
競合リスク入門
Noordzij M et al: Nephrol Dial Transplant.
2013 28(11):2670-7.
81