QUEST-KRITERIEN EINFÜHRUNGSVERANSTALTUNG CSP 2018 - Berliner Institut für Gesundheitsforschung 13. September 2018 - Berliner Institut für ...
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
EINFÜHRUNGSVERANSTALTUNG CSP 2018 QUEST-KRITERIEN Berliner Institut für Gesundheitsforschung 13. September 2018
Die folgenden Folien sind als Anregung zu verstehen, sie ersetzen nicht die AKTIVE Auseinandersetzung mit den QUEST-Kriterien im Hinblick auf Ihre spezifische Forschungsfrage. Bei Fragen wenden Sie sich gerne an Dr. Miriam Kip (miriam.kip@bihealth.de oder quest@bihealth.de). QUEST Tool box: (https://www.bihealth.org/en/quest-center/mission-approaches/englische- uebersetzung/the-quest-toolbox/)
MERIT-Project Development of attributes of a robust and innovative research/ Merkmale einer robusten und innovativen Miriam Kip Forschung (MERIT) QUEST-criteria: Axel Pries open questions: • priority setting • strategies of scientific rigor • transparency and dissemination or results • participation • intramural funding schemes, e.g. CSP, Validation fund • introduction to the doctoral/dissertation program at Charité
Warum? • Vorhandene Evidenz darlegen • Wissenslücken identifizieren • Bisherige Studienqualität kritisch bewerten • Identifikation von Faktoren, die die Effektivität einer Maßnahme beeinflussen • Ableiten wichtiger Informationen hinsichtlich Design neuer Studien • “evidence-based trial design” - Reduce waste in future research - Reduce risk for humans and animals - Reduce risk of unnecessary enrollment of humans
Wie? (a very short introduction) • Clinical Interventions - PICOS • Population Meshterms Volltextsuche • Intervention general expressions • Control/Comparators Boolsche Operanden • Outcome • Study design Filter • preclinical • (a) treatment/intervention • (b) disease or condition of interest • (c) animal species/cell population studied • (d) outcome measures
Wo? • PubMed, Embase • Clinicaltrials.gov, Metaregister der WHO etc.. • Eigene Vorstudien (dabei Daten nachvollziehbar darstellen)
Ressourcen (Auswahl) • BMJ 2011;343:d5928 doi: 10.1136/bmj.d5928 • Hooijmans et al. BMC Medical Research Methodology 2014, 14:43 http://www.biomedcentral.com/1471-2288/14/43 • http://syrf.org.uk/library/ • https://www.york.ac.uk/media/crd/Systematic_Reviews.pdf • https://www.cochrane.de/de/ressourcen • https://www.radboudumc.nl/en/research/radboud-technology- centers/animal-research-facility/systematic-review-center-for-laboratory- animal-experimentation • https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3265183/pdf/LA-11-087.pdf • https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3104815/pdf/LA-09-117.pdf
Phase III studies show no effect Minnerup et al. Exp.Transl. Stroke Med. 2014;6:2
Possible Solutions Reduce Bias! Use blinding, randomization,in/exclusion criteria. Report results according to guidelines (e.g. ARRIVE, CONSORT, PRISMA, etc.) Pre-register Use statistics sensibly! Go beyond the bar plot! Show individual data points and distributions. Think biological significance, think effect size. Consult a statistician.
Bias influences effect size Stroke models (NXY-095) Alzheimer's disease models models Improvement in behavioural outcome 1.2 (Standardised Effect Size) 1.0 Reduction in infarct size Reduction in infarct size 0.8 0.6 0.4 0.2 0.0 Yes No Blinded conduct of Blinded assessment Blinded assessment of experiment of outcome outcome > 30 studies > 500 animals
Tools and Resources Statistical Consulting (CRU and Biostats) QUEST toolbox (includes tools for Figure creation) Experimental Design Assistant for animal experiments Courses at the QUEST Center and Promotionskolleg
STRATEGIES OF SCIENTIFIC RIGOR (II)
Clinical Scientist Prof. Dr. Geraldine Rauch Dr. Jochen Kruppa Institute of Biometry and Clinical Epidemiology geraldine.rauch@charite.de UNIVERSITÄTSMEDIZIN BERLIN
Auswahl von Endpunkten Endpunkte: Diejenigen Merkmale in klinischen Studien, anhand derer der Erfolg der Studie gemessen werden soll. Man unterscheidet dabei zwischen primären und sekundären Endpunkten. Primäre Endpunkt: Erfasst dabei das Hauptziel der Studie. Dieses wird am Ende der Studie mit einem statistischen Test überprüft. Sekundären Endpunkte: Erfassen dabei weitere Aspekte der Studie. Die Auswertung erfolgt rein deskriptiv. Bei der Auswahl von Endpunkten sollte Folgendes beachtet werden: Angemessen für die medizinische Fragestellung Möglichst objektiv erfassbar Möglichst hohes Skalenniveau (vgl. spätere Folien)
Begriffe Zielgröße / Endpunkt: das Merkmal, das als Ergebnis einer Untersuchung interessiert, z.B. eine unter dem Einfluss der Therapie sich verändernder Laborwert oder ein Krankheitssymptom Einflussgrößen: alle Merkmale die im funktionellen Zusammenhang zur Zielgröße stehen, z.B. bestimmte Behandlungsformen, Therapiemaßnahmen Störgrößen/Confounder: Einflussgrößen, deren Untersuchung nicht Ziel der Studie ist, z.B. die unerwünschte Abhängigkeit vom Alter oder Geschlecht. Störgrößen sollten entweder eliminiert oder in der Analyse der Zielgrößen berücksichtigt werden. Störgrößen sind jedoch nicht immer alle erfassbar.
Harte und weiche Endpunkte • Harte Endpunkte – lassen sich direkt erheben – Beispiele: • Überlebenszeit • Tumoransprechen • Weiche Endpunkte – lassen sich nur indirekt erheben – Beispiele: • Lebensqualität • Schmerzempfinden
Studiendesign nicht kontrolliert/ kontrolliert/ einarmig mehrarmig Nicht für jede Fragestellung ist jedes Design möglich! nicht randomisiert randomisiert einfach offen verblindet doppelt Goldstandard für Studien monozentrisch multizentrisch zum Wirksamkeitsnachweis
Studientypen Es gibt viele verschiedene Studientypen Klassifizierung nach unterschiedlichen Kriterien möglich: • Fragestellung (z.B. Therapiestudie, Diagnosestudie, Prognosestudie), • Blickrichtung (prospektiv, retrospektiv), • “Aktivität” des Forschers – Beobachtungsstudien (Querschnittsstudien, Kohortenstudien, Fall-Kontroll-Studien) – Interventionsstudien (nichtrandomisierte Studien, randomisierte Studie) • Bei Arzneimittelstudien: „Phase“ der Erprobung (Phase I, II, III, IV) • usw. Hier kein umfassender Überblick möglich!
Quelle: Dtsch Arztebl 2009; 106(15)
Literaturempfehlung • Schumacher M, Schulgen G (2008, 3. Auflage): Methodik klinischer Studien, Springer. • Sachs L (1993, 7. Auflage): Statistische Methoden: Planung und Auswertung, Springer. • Weiß C (2010, 5. Auflage): Basiswissen Medizinische Statistik, Springer. • Gonick L (1993): The Cartoon Guide to Statsitics, HarperCollins Publisher
STRATEGIES OF SCIENTIFIC RIGOR (III)
Reveal, Don’t Conceal: Transforming Data Visualization to Improve Transparency Tracey L. Weissgerber, PhD Twitter: @T_Weissgerber
Data presentation is the foundation of our collective scientific knowledge… Figures are especially important. They often show data for key findings.
Many different data distributions can lead to the same bar graph… Symmetric Outlier Bimodal Unequal n 30 20 10 0 Test p value T-test: equal var. 0.035 0.074 0.033 0.051 T-test: Unequal var. 0.035 0.076 0.033 0.035 Wilcoxon 0.056 0.10 0.173 0.067 Weissgerber et al., PLOS Biology 2015
Why you shouldn’t use a bar graph even if your data are normally distributed Bar graph Bar graph Univariate (mean ± SE) with points scatterplot Zone of Range of Invisibility Observed Heart rate (beats/min) Values Zone of Irrelevance 0 Sedentary Exercise Sedentary Exercise Trained Trained Bar graphs 1. Don’t allow you to critically evaluate continuous data 2. Arbitrarily assign importance to bar height, instead of showing how the difference between means compares to the variability Weissgerber et al., JBC 2017
Graphics for: - Cross sectional studies - Experimental studies with independent groups Dotplot Boxplot with Boxplot Violin plot Bar graph points (with or without points) Outcome Continuous Continuous Continuous Continuous Counts & variable proportions Sample size Small Medium Large Medium to Any Large Data Any Any Do not use for Any N/A distribution bimodal data Free violin plot tool: https://interactive-graphics.shinyapps.io/violin/
Free Tools for Interactive Graphics 20 Group 1 Group 2 15 Group 3 10 5 0 Condition 1 Condition 2 Condition 3 Dot, box or violin plot: http://statistika.mfub.bg.ac.rs/interactive- dotplot/ Interactive line graph: http://statistika.mfub.bg.ac.rs/interactive-graph/ Additional resources: Twitter @T_Weissgerber
Why we need to report more than “Data were analyzed by t-tests and ANOVA” • Meta-research studies show that statistical errors are common. These include: – Failing to specify what test was used – Using suboptimal or inappropriate tests – Incorrect p-values • T-tests and ANOVA are the most common analysis techniques in many basic biomedical science fields
Why we need to report more than “Data were analyzed by t-tests and ANOVA” Systematic review: Many physiology papers are missing information needed to determine what type of ANOVA was performed Essential Details Papers with Missing Information Number of factors 17% Names of factors 54% Post-hoc tests 27% Between vs. within-subjects factors for 63% repeated measures ANOVA
Papers rarely contain information needed to verify the test result Essential Details Papers with Missing Information T-tests ANOVA (n = 163) (n = 225) Test statistic 96% 95% Degrees of freedom * 7% 97% Exact p-value 69% 78% * Exact sample size is also acceptable for t-tests
This information is essential to identify bias & correct errors 1. Confirm that the correct test was used 2. Confirm test results – Errors in reported p-value are common; may alter conclusions in 1/8 papers1 3. Assess bias: Were observations excluded without explanation? – Among papers with animal models of cancer & stroke2 • 7-8% excluded animals without explanation • 2/3 didn’t have enough information to assess 1Nuitjen et al., Behav Res Methods 2015 2Holman et al., PLOS Biol 2016
Solutions • Report exactly what test you used Test Reporting T-tests Unpaired vs. paired, equal vs. unequal variance ANOVA Number & names of factors, between vs. within subjects factors, post-hoc tests, any interaction terms included in the ANOVA More complex tests Detail needed to reproduce analysis
Why does this matter? An example… These two ANOVA Repeated Measures tests… without repeated measures ANOVA …see the data Compares 60 Compares 60 differently 3 independent groups 40 10 pairs 40 (n = 30, 10/group) of related 20 20 observations 0 (n = 10) 0 T1 T2 T3 0 T1 1 T2 2 T3 3 …test different Null hypothesis: Null hypothesis: hypotheses Mean T1 = Mean T2 = Mean T3 Mean T1 = Mean T2 = Mean T3 when population means are related …use More unexplained variability Less unexplained variability – we can information account for the effect of “subject”. differently Sums of squares Sums of squares Conditions (between groups) 240 Conditions (between groups) 240 Residual (within groups) 2604 Residual (within groups) 2604 Subjects 2195 ? Error 409 Total 2844 Total 2844 …give different overall p = 0.304 overall p = 0.016 results
Solutions • Report exactly what test you used • Improve clarity by describing simple tests in figure & table legends • Report test statistic, degrees of freedom, exact p-value • Use Statcheck: http://statcheck.io • Deposit code: Make your analyses reproducible Clear reporting allows you to identify & correct errors prior to publication!
STAKEHOLDER ENGAGEMENT
Patient Engagement in biomedical research: From research subjects to partners in research Miravittles M et al. (2013), Respiratory Medicine 107, 1977-1985 Engaging patients as partners and not just subjects in research can improve research! They hold important experience-based expertise from living with diseases.
Patient Engagement in biomedical research: Key areas for patient engagement Geissler J et al. (2017), Therapeutic Innovation & Regulatory Science 51(5), 612-619. Key areas Research phases Key areas
Patient Engagement in biomedical research: Important resources Possible gatekeeper to Best Practice Examples identify patients… …although so far no institution appointed.
Patient Engagement in biomedical research: Who else should be involved? Other stakeholders whose involvement might benefit research: • Political actors involved in decision-making about medical products, e.g. G-BA, IQWiG, BfArM, PEI? • Industry actors from the pharmaceutical and biotech world? • Clinicians using the research outcomes? • Training institutions teaching the findings? • …
Patient Engagement in biomedical research: Further open questions • Who should represent (and why) particular stakeholder groups? • Do representatives need particular training? • How should engagement activities be structured to ensure a level playing field? • What conflicts of interest exist and how can they be managed? • Is there a need for a coordinating institution for patient engagement? • …
You can also read