1、Medical Research paper writing,Chao Mi,Medical Research paper writing,Unit 1 Planning and Observing,LOGO,Medical Research paper writing,Literature Search Use data base (MEDLAR/MEDLINE) Focus on study design (is your superior to others) Avoid derivative/secondary work Team building Talk with previous

2、 research teams A well-known investigator (blinded system) Full time research nurses (detail/objective) A talented writer A qualified statistician Methodology Stating the problem (the purpose of your study),LOGO,Medical Research paper writing,e.g. Approximately 40% of patients with spinal cord injur

3、y develop a pressure ulcer during the initial hospitalization. However, there is currently no method of quantifying this risk that is accurate for this population. Formulating the hypothesis: Start with H0 that there are no connections with the two variables (smoking history is not associated with t

4、he incidence of lung cancer) Ha is the opposite of H0; But use it with H0 at least. Study design: Beware this is the most common type of flaw that results in outright rejection of a manuscript! Categories (Observational/Quasi-experimental/Experimental),LOGO,Reading,Examples: 1. Researchers compare n

5、ewborn infants from two groups. In one group, the mothers used cocaine during pregnancy. In the other group, the mothers did not use cocaine during pregnancy. 2. Researchers compare the outcome of infants born in two cities. In one city, a new law was enacted to test pregnant women for cocaine use a

6、nd enroll those who had positive test results in a treatment program. The other city had no such law or program. 3. Researchers compare newborn rats from two groups. In one group, the researchers had given the mothers cocaine during pregnancy. The mothers of the rats in the other group were not give

7、n cocaine during pregnancy.,LOGO,Reading,handout material reading (Case-control; Cohort; Historical perspective; Cross-sectional) Minimizing Bias: Selection bias: a systematic difference between people in the study and people not selected. (same area; long period and follow-up information) Response

8、bias: respondents differ systematically from non respondents. (people with certain disease may tend to respond to a mailed questionnaire) Information bias: difference between measurements in different study groups. (patients with risk factors may be tested more carefully and frequently),LOGO,Reading

9、,Confounding bias: effect of extraneous variables that must be adjusted before research questions are answered. An inner-city hospital may have a higher neonatal mortality rate than the national average. So the confounding variables associated with inner-city life may create a confounding bias. Data

10、 Collection: Use questions with defined choices not open-ended questions and cover all possibilities. Categorical data is inferior to continuous data (bell curve) Dummy questions should be used.,LOGO,Reading,Failure to collect data on variables that could influence the Interpretation of results. (ou

11、tright rejection of an article So use others experience) Reliability and validity: Reliability: measurement used on the same patient under the same condition would yield the same results. e.g. Low reliability: self-reported duration of exposure to passive smoking among women breastfeeding their infa

12、nt. e.g. High reliability: nicotine concentration in the mothers breast milk. Validity: Can a test or procedure measure well. e.g. Malnutrition could be assessed by asking patients to score their nutritional intake on a scale of 1 to 10, but measuring their serum albumin level is more valid,LOGO,Rea

13、ding,Eligibility(適合性) Patients should be homogeneous (including a mixture of diseases or procedures) If you use a mixed sample, pay attention to: Inclusion and exclusion decisions; Distribution of disease severity of the patients and why; statistical adjustment for differences among these patients C

14、ombining different types of poor outcome because each one is small. Use the outcome that could be predictable by the independent variables and explain why. All diagnoses and procedures must be encoded for analysis (ICD; The AMA CPT Code Book): e.g. In diagnosis fields, hip fracture is represented by

15、 ICD code from 820.00 to 820.99 (use standard),LOGO,Reading,Randomization and Confidentiality Strengthen the study design and minimize bias. Randomized clinical trial: patients are randomly assigned to either receive or not receive the intervention. Double-blinded design: clinician not knowing the a

16、ctive treatment and placebo. Triple-blinded design: even statistician not knowing who receive active treatment and placebo. e.g. A study is being planned in which four research nurses review the hospital charts of trauma patients who are admitted to 10 hospitals They know that two of the nurses are

17、much more aggressive in searching charts for complications. To minimize this potential bias, they use a random number table to assign monthly blocks of charts to be reviewed at the 10 hospital.,LOGO,Reading,LOGO,Although this study is not a randomized clinical trial, the use of randomization strengt

18、hens the study design and reduces bias. Protect the confidentiality of all participants. Abide by the documentation of IRB (Institutional Review Board); Patient, hospital, physician, ect (case number 1, 2, 3) End points and Outcome Units of analysis e.g. Evaluating the accuracy of screening for feta

19、l problems. (1) a pregnant woman; (2) an infant; (3) a pregnancy; (4) a singleton pregnancy that continues beyond 28 weeks of gestation Confounding factors All the other things that could explain your results. From crude rates to adjusted rates (statistical transformation has been done to eliminate

20、differences between groups),Reading,What is case mix or patient mix? e.g. Compare survival rates among hospitals for patients who underwent a specific surgical procedure, crude survival rates are almost meaningless. The investigator must adjust for confounding factors, such as differences in patient

21、s (age), preexisting conditions (diabetes), or whether the procedure was performed on an emergency or an elective basis. These factors are referred to Locate confounding factors Perform a literature search to identify them before data collection is finished. Classification of variables. continuous (

22、bell curve; hematocrit 血細胞比容) categorical (groups; anemia)/dichotomous (just 2 groups; alive vs dead),LOGO,Reading,An example of recoding and combining data: Length of stay in the hospital is often a skewed continuous dependent variable. So set 0 for patients whose stay is less than the median and s

23、et to 1 for the other patients. A much smaller sample, but statistical techniques are better applied. Data entry Always quantify! Trick: use interval scale when possible. Avoid mixing text and code. Use big number for Missing data or Unknown. (9, -99, not NA) Logical coding: 0=no, 1=yes; Illogical c

24、oding: 1=none, 2=none, 3=two,LOGO,Reading,Examples of weak and strong variables: 1. Low Birth Weight Exact Birth Weight 2. Respiratory distress syndrome Lung index 3. NICU admission Total cost of hospitalization or days in the NICU (新生兒重癥監(jiān)護室 P38) Sample Size Estimating Sample Size In the planning ph

25、ase, review sample size in similar published study (consider both control and subgroup) Statistical Power The probability of rejecting the null where it is false. Taken with P value, sample size is determined by using Table on page43.,LOGO,Reading,Preparing for Statistical Analysis Levels of measure

26、ment 1. Nominal: 0=no, 1=yes; 1=female, 2=male; 1=white, 2=black, 3=Asian 2. Ordinal: 0=nonsmoker, 1=light smoker, 2=moderate smoker, 3=heavy smoker 3. Interval: Age, Hematocrit, Serum albumin level, cigar smoked per day Planning the following-up 1. Impressive length of follow-up (25 years) 2. Avera

27、ge time may differ between the treatment and control group.,LOGO,Reading,3. Intention-to-treat analysis: data from all patients must be included. (not only who finished the study, e.g actual on therapy analysis) Avoiding Common Criticisms (peers/editors) Design of the study Poor experimental design/

28、methodology No accounting for confounding factors/ biased data small sample size/ improper statistical methods Interpretation of findings Unfounded/ unsupported findings Over interpretation of data Inadequate discussion/ unexplained inconsistencies,LOGO,Reading,Importance of the topic Rehash of esta

29、blished facts/ unimportant topic No reader interest/ not generalizable Little clinical relevance Presentation of the results Poorly focused/ organized/ written Too long/ poor grammar Introduction and Discussion tend to be too long Methods and Results tend to be too short Collecting Data Periodic mon

30、itoring of the studys progress Keeping records,LOGO,Reading,Notes that includes detailed research decisions Data coding conventions and abstracting guides Data collection form Reference research paper The instruction for author from target journal One recent article from target journal e.g. Both sta

31、tistical and clinical implications of changes must be considered properly. Number of, reason for, a missing patient. Detecting potential problems Distinguish between missing data and the answer “no” Monitoring the sample size reevaluate the sample size with new realistic estimates.,LOGO,Reading,Stat

32、istical Analysis Laying the groundwork Build a database for statistical analysis; Use logical, easy to remember field name; Dont ask inexperienced programmer or medical student to customize a database. e.g. Fixed field Format File CASE GROUP BPs SEX RACE AGE LOS ICU 1 3 120 1 1 24.6 22 9 2 2 80 2 3

33、0.5 15 1 3 3 130 1 5 -99.9 10 0,LOGO,Reading,Preparing the Data “Clean” and “freeze” the data 1 Avoid different units (week/month/year) 2 Complete the missing code (as unknown/ drop the case when essential data is missing after discussion) 3 Document the change made in the data 4 Freezing means no d

34、ata will be added or changed during next phase of analysis. Interpreting the data Preparing for data analysis Do not force numbers to prove your point. Cannot simply press a button and let PC analyze your data Consult a data analyst with adequate experience,LOGO,Reading,Create equal-sized groups of

35、patients for continuous variables e.g. You may not have enough data for certain subgroups to carry out analysis. Grouping of patients by age decades (0-9, 10-19, 20-29) might be replaced by deciles (0-6.4, 6.5-14.9, 15.0-22.7) Avoid common problems Dont forget to repeat your experiments. Control gro

36、ups should not be poorly matched. Epidemiologic techniques can be used to strengthen clinical research. For examples: 1. Person-years: the number of years that each member of a population has been afflicted by a certain condition. (drug use) 2. Survival analysis and follow-up life tables: can be use

37、d to compare groups that are followed up for varying periods.,LOGO,Reading,Dose-response relationship is a change in the amount, intensity, or duration of exposure associated with a change in the risk of a specified outcome. e.g. The incidence of lung cancer increases with the number of cigarettes s

38、moked per day. Univariate Analysis Most common univariate tests used Chi-square test and Students T-test (simplistic but essential first step) Using Chi-square test Test categorical variables (Whether the actual proportion in the groups differs significantly from the proportion expected by chance al

39、one),LOGO,Reading,Using Students T test This test can help detect whether the means for the two unmatched groups are significantly different. (Bell shaped data) When Students T test is not preferred: a. More than 2 groups (one-way analysis of variance: ANOVA) b. Data is not bell- shaped (nonparametr

40、ic test) c. Skewed data (2 groups/ Mann-Whitney U test) d. Skewed data (2 groups/ Krushal-Wallis test) Sometimes more than one tests are used together, but never use the improper one to fish for significance, which will lead to false conclusions. Nonparametric test Data not normally distributed or o

41、rdinal 2 groups: Mann-Whitney U test; 2 groups: Krushal-Wallis test,LOGO,Reading,2 groups (matched): Wilcoxon signed-rank test; 2 groups (matched) : Friedman ANOVA by ranks Matching: Purpose: An efficient way of creating a control group based on confounding factors. (Best before data is collected) e

42、.g. Preventable deaths in trauma care: statistically controlling for the severity of injury within each specific mechanism of injury. (Actually adjusting both severity and mechanism is not realistic; and even this causes problems; P87) Multivariate Analysis When to use? One dependent variable vs mul

43、tiple independent variables. But look at the following example:,LOGO,Reading,e.g. Compare the infection rate of patients receiving drug A with that of patients receiving drug B. After randomizing the patients into two groups, you find one group is significantly older than the other. Now you can use

44、multivariate analysis to adjust for the age of the patients So use multivariate analysis to deal with confounding factors and do mention this is done in your paper. For this, you can refer to research papers in your field for variables that are potentially confounding factors. Logistic regression an

45、alysis Examine the independent contribution of more than one predictor variable. (The logistic regression model showed that among mothers who smoked cigarettes during their pregnancy, the infant outcome can not be predicted by the risk factor under study.),LOGO,Reading,Give it a second thought befor

46、e conclude: there are no difference between the study group (Interaction) e.g. Investigators found that among people with spinal cord injury, both diabetes and smoking were moderate risk factors for pressure ulcers. However, all of the patients who had diabetes and also smoked had a history of press

47、ure ulcers. This effect from a combination of factors is an interaction. Always translate multivariate analysis into plain English because reviewers and editors are not statistical experts. Present the results of multivatiate analysis in a clinically useful format. Use evidence to support each step

48、of the analysis. (Use specific data to just prove your model- bad!),LOGO,Reading,P value is not the only standard; consider what makes sense clinically. (preexisting diseases; injury severity; time; treatment) Learn Cox Proportional-Hazards Regression to Publish Papers in a High-Quality Journal.,LOG

49、O,Reading,Unit 2 Title and Abstract (1),LOGO,Title,Principles: 1 Accuracy 準確 2 Brevity 簡潔 3 Clarity 清楚 4 Effectiveness 有效 5 Attractive 吸引 Mistake 1: too big or too small e.g. Transplantation of corneal stem cells cultured on amniotic membrane for corneal burn: experimental and clinical study (CMJ,20

50、02,115(5):767-769) experimental and clinical - in vitro and in vivo (too big),LOGO,Title,Mistake 2: not accurate e.g. Experimental study on mechanism and rarity of metastases in skeletal muscle. (CMJ, 2002, 115(11) 沒有把研究對象、問題說清楚。 Better no more than 100 characters (title + subtitle) A good example:

51、Importance of replication in microarray gene expression studies: statistical method and evidence from repetitive cDNA hybridization (Proc Natl Acad Sci USA) Attractiveness (醒目) e.g. Identification of human tumor initiating cells Nature, 2004-11-18, 432(7015): 396-401 對比：Mechanism of ligustrazini aga

52、inst thrombosis (CMJ) Too general,LOGO,Reading,兩個省略不當?shù)睦樱?1. Adhesion kinase controls actin assembly 2. Mouse behavior,LOGO,Reading,Correct: Adhesion kinase controls actin assembly via a FERM-mediated interaction with Arp2/3 complex （介詞） Androgen Administration to aged Male Mice Increases Anti-Anxie

53、ty Behavior and Enhances Cognitive Performance (verb),LOGO,Reading,幾點注意事項： 1. Plain English 2. No underline or Italics 3. Only Standard Abbr 4. Capitalize the initial letter of each word 5. Seldom use “The Study of” or “The Research on” e.g. The study of Argonaute2 in mammalian gastrulation and prop

54、er mesoderm formation Argonaute2 is essential for mammalian gastrulation and proper mesoderm formation,LOGO,Reading,句法結構: 詞序非常重要； 多由名詞性短語構成； 動名詞或分詞；少用 陳述句。 如：你想表達B7-H4蛋白主要表達在非增生性腫瘤細胞中，其中 一部分為腫瘤肝細胞： 修改前： B7-H4 most expresses on Ki67 negative brian tumor cells and can express on some brain tumor stem

55、cells. 修改后：,LOGO,Reading,句法結構: 詞序非常重要； 多由名詞性短語構成； 動名詞或分詞；少用 陳述句。 修改后： B7-H4 is predominantly restricted to Ki67- negative brain tumor including a small subpopulation of brain tumor stem-like cells.,LOGO,Reading,結構類型 （國內較高水平級期刊）： 1. 短語型: (1) 名詞短語： a. 名詞+介詞短語 用食管鏡檢查以預計曲張靜脈出血 b. 定語+名詞 原發(fā)性肝癌二級預防的隨機分組前沿性

56、研究 c. 名詞+分詞短語 何杰金病放射治療引起的甲狀腺癌,LOGO,Reading,結構類型 （國內較高水平級期刊）： 1. 短語型: (1) 名詞短語： a. 名詞+介詞短語 用食管鏡檢查以預計曲張靜脈出血 Prediction of varicose haemorrhage by esophageal endoscopy b. 定語+名詞 原發(fā)性肝癌二級預防的隨機分組前沿性研究 Randomized controlled prospective study of secondary prevention for primary liver cancer c. 名詞+分詞短語 何杰金病放射

57、治療引起的甲狀腺癌 Thyroid carcinoma induced by irradiation for Hodgkins disease,LOGO,Reading,(2) 動名詞短語： 心房纖顫患者中風預防 (3) 介詞短語： 原發(fā)性肝癌臨床診斷探討 句子型： （1）陳述句： 人類hR24L基因參與DNA切除修復和重組修復的研究 （2）疑問句 （常見于評述性文章）： 尚未解決的問題 飲酒者冠心病少嗎？,LOGO,Reading,(2) 動名詞短語： 心房纖顫患者中風預防 Preventing stroke in patients with atrial fibrillation (3)

58、介詞短語： 原發(fā)性肝癌臨床診斷探討 On clinical diagnosis of primary carcinoma of liver 句子型： （1）陳述句： 人類hR24L基因參與DNA切除修復和重組修復的研究 Human hR24L gene is involved in DNA excision repair and recombination repair. （2）疑問句 （常見于評述性文章）： 尚未解決的問題 飲酒者冠心病少嗎？,LOGO,Reading,Unresolved issue: Do drinkers have less coronary heart disease? 中心詞的使用：（一般放在句首） e.g. 針療對冠心病心絞痛患者左心功能狀態(tài)的影響 e.g. 氟烷麻醉治療新生兒受損肺功能,LOGO,Reading,Unresolved issue: Do drinkers have less coronary heart disease? 中心詞的使用：（一般放在句首） e.g. 針療對冠心病心絞痛患者左心功能狀態(tài)的影響 The acupuncture effect on left ventricular fun