Public Articles
История
Начальная история Руси до конца X века.
Происхождение восточных славян.
Социально - политическая ситуация славян.
Международное положение восточнославянских земель.
Образование древнерусского государства
BACA methods
Habits that I want to bring back from Burning Man
What Happened to Distributed Programming Languages?
Nowadays, most programs we write are in some sense distributed—making HTTP requests or serving responses over HTTP, fetching or computing data on some remote resource, building some microservice that is meant to interact with others, etc. With all of this distribution going on, one might ask, what happened to distributed programming languages? Why are we still using languages like Java or C++ for these sorts of tasks? In this talk, Heather will take us on a whirlwind tour through history up to the present of distributed programming languages as well as programming constructs meant for distribution like futures and RPC. Together, we'll try to work out what happened to all of the distributed programming languages!
BeerDeCoded: preliminary results of the open beer metagenome project
and 3 collaborators
How to conduct meta analysis and use GRADEPro for best evidence
Meta analysis refers to a process of combining the results of primary studies to arrive at a pooled estimate of the that estimate. For example, let's say you are interested to study if a particular drug X is effective in the treatment of high blood pressure. For this, you can either review the results of a single trial (a randomised controlled trial), or you can pool the results of more than one randomised controlled trial for similar patients and arrive at a summary estimate of the effect as to whether the drug X is beneficial for high blood pressure; also, based on the subset of studies, you can find out if the harms associated with the drug X is large enough when the resuls of the studies are pooled. In this context, I wrote about an intervention and effect of that intervention on an outcome, but you can also consider effect of an exposure (as happens in observational epidemiological studies); in this way, you can conduct a meta analysis of any number (as long as that number is two or more) of cohort or case control studies to test the association between an exposure and an outcome variable. For example, if you want to study the effect of smoking on heart disease, you can conduct a meta analysis to test the effect estimate of smoking on heart disease by pooling together results of case control studies on the association between smoking and heart disease.
In this part (part I), we are going to learn how we can use GRADE and R to conduct a meta analysis. In the next part, we will introduce GRADE and GRADEPro software for conducting an evidence appraisal in another way. R is a free and open source software you can use for statistical programming and statistical data analysis. In this part, we are only going to use a small set of functions with R. You will not need to know the nitty gritty of working with R in order to work through this module (although I recommend that you study that); you can work through the codes by selecting and running them in the RStudio server instance we have set up for you. We will conduct a meta analysis and we will use GRADE approach and GRADEPro software to learn how you can approach quality assessment of studies using GRADE (just the principles and some applications). How to use GRADE will be taken up in the next module. GRADE is a short hand for grading, recommendations, appraisal, development and evaluation of guidelines. While it is used for guideline (clinical and public health practice) guidelines development, you can also use GRADE for other purposes. Meta analysis is based on appraisal of individual studies, GRADE is used for appraisal of quality of individual outcomes. You can use meta analysis within a GRADE process.
We have learned that evidence based practice consists of a series of steps where we identify a problem, we set up a series of questions (one or more questions), we search for relevant evidence, we appraise that evidence for possible biases in the research, and we identify the results of the studies. We have also learned that in evidence based healthcare, if we say that X causes Y (that is, if X as an intervention can cause Y a health outcome, or if Y a health outcome is causally associated with X an exposure), then X must satisfy the following nine criteria (not all of them, but at least the first four criteria of the following):
First, we must rule out any play of chance in the association between X and Y. We ensure in a study setting by selecting a proper sample size and sufficient power;
Second, we must make sure that the study or studies we are reviewing will have little bias; if the study has significant biases either in the allocation of exposure or intervention or observation of effects, then this bias will significantly compromise with the internal validity of such a study/studies
Third, we must make sure that all possible confounding variables have been taken care of, or as many possible confounding variables as possible were taken care of; otherwise, this too, is a factor that will interfere with the internal validity of the study. Normally, confounding is taken care of either in the design of the study (randomisation by design take care of observed and unobserved confounders; matching in case control studies take care of this only for those variables that are thought to be confounders ), or in the analysis phase where researchers use multivariate analyses;
Fourth, if X is to ba cause of Y, then X must precede Y in time sequence, so X must come first and then Y.
Other than these restrictions, in 1965, Sir Austin Bradford Hill discussed nine considerations that included the time sequence and these are used to ascertain if the nature is one of cause and effect (cite Hill).
We also know that in a hierarchy of evidence in terms of what study design is best suited for framing high quality evidence and what study design is considered to be not so high quality of evidence, we have almost a ladder of study designs. At the top of that hierarchy we have meta analyses of randomised controlled trials, and at the bottom rung of that ladder, we have case studies and case series. We have learned that RCTs are high quality study designs because they take care of selection bias and confounders, and when blinded (the more blinding the better), they are excellent study designs to take care of most biases and all confounding variables. Besides, if the RCTs are large in size (with many individuals enrolled), then these are good study designs to bank on.
In meta analyses, we will pool the results of similar RCTS for studying the health effects of specific interventions. For this exercise, we have selected the following problem.
Neuropathic pain is caused due to damaged nerves and the type of pain is usually treated with opioids. Morphine is an opioid that is used to treat neuropathic pain. What is the effectiveness of morphine in treating neuropathic pain in adults in pain relief and how safe is it?
We will use GRADE and RStudio simultaneously for this exercise. Our meta analysis exercise will consist of the following 10 steps:
From the above problem, we set up the following PICO formatted question:
"Among adults with neuropathic pain, what is the effectiveness of morphine compared with placebo in relieving pain to the extent of at least 30% from baseline and little adverse effects?"
The PICO are as follows:
"P": "Adults with neuropathic pain"
"I": "Morphine"
"C": "Placebo or other treatments"
"O": 30% or more pain relief, or patient reported substantial change (patient global impression of change or PGIC)
This question will help us to set up selection/rejection criteria and terms for our search of litearature databases:
We will use the following search algorithm on Medline/Pubmed. Open Pubmed webpage and view the page that should look as follows:
We will reproduce the following search strategy there to get started: (download the following text and type the search terms in the pubmed box)
If you followed the search terms in the table and typed them in Pubmed, you will end up with 3 articles.
This was reproduced from another Cochrane Meta-analysis and they searched different other databases, so we will not go into the details of those searches. The Cochrane review is linked here and you can go over the trial details.
Based on the search terms and selection process, the authors identified five studies. These studies are:
Gilron I, Bailey JM, Dongsheng T, Holden RR, Weaver DF, Houlden RL. Morphine, gabapentin, or their combination for neuropathic pain. New England Journal of Medicine 2005;352(13):1324-34
[DOI: 10.1056/NEJMoa042580]
Gilron I, Tu D, Holden RR, Jackson AC, DuMerton-Shore D. Combination of morphine with nortriptyline for neuropathic pain. Pain 2015;156(8):1440-8.
[DOI: 10.1097/j.pain.0000000000000149]
Huse E, Larbig W, Flor H, Birbaumer N. The effect of opioids on phantom limb pain and cortical reorganization. Pain 2001;90(1-2):47-55.
[DOI: 10.1016/S0304-3959(00)00385-7]
Khoromi S, Cui L, Nackers L, Max MB. Morphine, nortriptyline and their combination vs. placebo in patients with chronic lumbar root pain. Pain 2007;130(1-2):66-75
[DOI: 10.1016/j.pain.2006.10.029]
Wu CL, Agarwal S, Tella PK, Klick B, Clark MR, Haythornthwaite JA, et al. Morphine versus mexiletine for treatment of postamputation pain. Anesthesiology 2008;109(2):289-96.
[DOI: 10.1097/ALN.0b013e31817f4523]
We will now use GRADEpro software to fill in the data from each of these articles and will learn how GRADE can be used to construct evidence. However, it is important to note that GRADE approach uses outcomes as units of analysis rather than individual studies as units of analysis as we do in meta analyses.
The first step is to visit the GRADEpro website and register yourself:
After you log in, click on "New project", when you do so, it will look like as follows:
Fill in as shown in this figure and select GRADE Evidence Profile from the drop down box. Then click "Create Project"; you can do different things in the GRADEpro, and a full description of everything that is possible in GRADE pro is out of the scope of this class or tutorial. Instead, now focus on three boxes displayed. Click on "Add management question" because we are interested in finding out the effectiveness of morphine for pain relief. This will open a screen as follows and fill in the relevant boxes. When you do so, the screen should look like as follows:
You do not have to fill in the bibliography or the question authors at this stage. You have only one problem statement and one question here. So, if you click on the problem statement on the top of this box, this will open the following window for you:
Click on "Add outcome" to proceed. We are going to use one study (Gilron I, Bailey JM, Dongsheng T, Holden RR, Weaver DF, Houlden RL. Morphine, gabapentin, or their combination for neuropathic pain. New England Journal of Medicine 2005;352(13):1324-34
[DOI: 10.1056/NEJMoa042580]) to fill in the details. You can do the other studies yourself. GRADE approach is about outcomes, not individual studies, so the purpose of this exercise is to show you how you can fill in the details. In future, we will see how we can fill in the details from multiple studies. So, after you click, add outcome and abstract information from the Gilron study, the page will look like as follows:
This is the first thing you do, that is, fill in the information about outcome measurement and give the outcome a name. Where in the article do we find out the information? You find that information in the section named "outcome measurement". If your article does not contain that information, look for it. For example, in this article, the information was contained in the following section:
Now, click on the "save" icon and this will take you to the screen where you will fill in the boxes. We will describe the processes, step by step:
Title of the Box | What to fill in and why? |
---|---|
Number of studies | Here, one, otherwise ... |
Study design | Randomized Trial (1) |
Risk of Bias | Not Serious (2) |
Inconsistency | Not Serious (3) |
Indirectness | Not Serious (4) |
Explanations:
Now we turn to the results section of the paper and fill in the remaining bits.
Condition | What decision |
---|---|
Imprecision | Not serious (1) |
Other considerations | None (2) |
When you look for imprecision, look for if the authors have reported p-values and 95% confidence intervals. If the p-value is less than 0.05, or if the 95% confidence interval range does not include the null value within it, then you know that the results are not imprecise. The null value for a difference will be "0", and null value for Odds Ratio or Relative Risk will be "1". Hence, all of the following will be imprecise:
A p-value of 0.20
An Odds Ratio of 1.5 (95% CI: 0.7-2.3)
A Relative Risk of 2.3 (95% CI: 0.8-3.5)
Now we move to the next part of the GRADEPro decision making:
There are two other columns, Quality and Importance will need to be worked out. GRADEpro automatically estimates the quality of either one study or the body of evidence; the importance is determined by you.
How many people were in the "Morphine" arm. We estimate that using the figures from the article below:
screening <- 86
excluded <- 29
withdrew <- 16
participated <- screening - (excluded + withdrew)
(participated)
Because these individuals received sequences of intervention and placebo, the intervention and the placebo arm were both based on 41 individuals. For the effect size where we are only going to compare 0-10 point pain scale, we see in the results section,
"Mean pain intensity on a scale from 0-10 at baseline and at the maximum tolerated dose was calcuated as follows: mean(+/- SE) at baseline, 5.72 +/- 0.23, with placebo 4.15 +/- 0.33 with Morphine" but a clear standard deviation is not available. Hence, the effect size:
baseline <- 5.72
final <- 4.15
effect_size <- final - baseline
(effect_size)
So, we see there is a "reduction" of 1.57 points on the pain scale; this should be entered in the boxes. Finally, we decide how critical is this outcome for our decision making. We decide that this was "critical". Hence, at the end of this exercise, our single study GRADE will look like as follows:
Now, let's learn how to conduct meta analysis in binary outcomes. We will abstract data from the meta analysis as shown here:
The authors of the meta analysis provide data for four studies: Gilron (2005), Huse(2001), Khoromi(2007), and Wu(2008). In each case the outcome was relief of pain 30% or more. Either the patients included in the study experienced such an improvement or they didn't hence this is a binary outcome. Here is a table
mydata <- read.csv("binarydata.csv") # read the data for meta analysis
mydata$mor_ne <- mydata$morphine_total - mydata$morphine_event # we create variables for non-event for morphine group
mydata$pl_ne <- mydata$placebo_total - mydata$placebo_event # we create variables for non-event for placebo group
The above codes help us to create the data sets that we will use for the binary meta analysis from the data we already have. Next, using R, we install and load the "meta" package of R. A package contains functions and data. See:
# install.packages("meta"dep = T,)
library("meta")
Now, we will use the "meta" package in R to conduct the meta analysis
mymeta <- metabin(event.e = morphine_event,
n.e = morphine_total,
event.c = placebo_event,
n.c = placebo_total,
data = mydata,
method = "MH") # Explain the codes
summary(mymeta) # summary of the meta analysis
forest.meta(mymeta) # draw the forest plot
funnel.meta(mymeta) # draw the funnel plot
This produces a meta analysis. First understand the output:
Then, the forest plot:
Finally to test the publication bias the funnel plot
La formación como herramienta de futuro
and 1 collaborator
Inteligencia Artificial Aplicada al Territorio
Data Mining Home Work 1 U95489771
n+(n-1)+(n-2)........
n+ (n-1) + (n-2) + ......... + 1 = 100n(n+1)/2 = 100 ( because, sum of first n natural number is given by n(n+1)/2)n=13.7
create two temporary arrays tempLeft and tempRight of size n complexity: {O(1) }create two variables productLeft and product Right and set them to 1
set tempLeft[i]= productLeft and set tempRight[ n-1]=productRightset productLeft=productLeft * V[i] and productRight=productRight * V[n-i]increment the variable iloop till the end of the array (i<n.)
Set tempLeft[j]=tempLeft[j]*tempRight[j]interate the variable jloop till the end of the array(j<n)
Untitled Document
Machine Learning for Classification of Disease-Causing Vectors
Data Mining Home Work 1
n+(n-1)+(n-2)........
n+ (n-1) + (n-2) + ......... + 1 = 100n(n+1)/2 = 100 ( because, sum of first n natural number is given by n(n+1)/2)n=13.7
Efficient Relative Solvation Free Energies with QM/MM
Kruskal Wallis Test or Multiple T test?
wet lab repeats papers
top STR genes
coor | gene_id | region | repeat_unit | pvalues | p.adjusted | trend |
1:16533831-16533854 | ENSG00000142632 | intronic | TTCA | 2.48E-38 | 1.25E-33 | increasing |
6:37662442-37662460 | ENSG00000112139 | intronic | CCT | 8.95E-27 | 2.25E-22 | increasing |
19:21712416-21712433 | ENSG00000197013 | intronic | CT | 1.01E-19 | 1.69E-15 | increasing |
22:42896554-42896575 | ENSG00000172250 | upstream | GGGCG | 6.73E-19 | 8.47E-15 | decreasing |
19:6434925-6434941 | ENSG00000181240 | upstream | ACA | 2.20E-17 | 2.21E-13 | increasing |
8:79598667-79598699 | ENSG00000104427 | intronic | TTAT | 9.64E-16 | 8.09E-12 | decreasing |
7:23670117-23670135 | ENSG00000169193 | intronic | CAAT | 9.33E-15 | 6.71E-11 | increasing |
17:80417741-80417765 | ENSG00000141562 | intronic | AAAA | 4.70E-14 | 2.96E-10 | increasing |
21:45052367-45052380 | ENSG00000160207 | intronic | ACA | 8.07E-14 | 4.51E-10 | increasing |
1:25617041-25617079 | ENSG00000187010 | intronic | GAAA | 2.14E-13 | 1.08E-09 | increasing |
What you need to know about: High flow nasal oxygen therapy
and 1 collaborator
Causal Inference Methods for Anaesthesia and Perioperative Medicine Research
and 1 collaborator
The introduction section will introduce the ideas behind Estimation vs. Identification in medical research, the Problem of Confounding, and Graphical Models (Directed Acyclic Graphs, or DAGs).
3 Reasons Why to Choose Emirates Airline
Ideen für Weihnachtsfeier
EDFN6020: Instrument Survey Document
and 2 collaborators
An individual's sense of their surrounding based on direct observation.
The atmosphere created by the interactions between students.
The set of beliefs, behaviors, attitudes, expressions, and dispositions shared by a set or subset of students
A tuition-free high school formed by a charter granted by a governing agency allowing the school to be managed indepently from the district-school board while still requiring the school to be accountable to all applicable educational law.
Community of people who will be understanding, supportive, and trustworthy someone seeks help, advice, or just someone to talk to. \cite{nokey_baeb4}
Teachers' belief that students identify as a larger commuinty and are members of that community.
Teachers' belief that the general student population and student sub-groups positively-integrate.
Teachers' belief that students' word-choice, language, and communication demonstrate respect towards peers.
Teachers' belief that the students view the school as a SafeZone.
Teachers' belief that students can influence their peers and be accepted by their peers while leading.
Teachers' belief that students engage in opportunities to make decisions, voice their opinions, and advocate for school-change.
Teachers' belief that students are self-accountable learner and hold all teachers to the same expectations.
Math Econ Ch. 20 Notes & Class
Maximize \(\int ^T _0 F(t, y, u) dt\)subject to \(\frac{fy}{dt} \equiv y' = f(t, y, u)\)and \(y(0) = A; y(T) = \) free\(u(t) \in U\); for all \( t \in [0, T]\)
Maximize \(\int ^T _0 U(C) dt\)subject to \(\frac{dK}{dt} = Y( K, L) - C\)and \(K(0) = K_0\) \(K(T) = K_T\)
\(H(t, y, u, \lambda) \equiv F(t, y, u) + \lambda(t) f(t, y, u)\)
\(y' = f(t, y, u) \Longrightarrow \frac{\partial H}{\partial \lambda}\)
\(\lambda ' (\equiv \frac{d \lambda}{dt}) = -\frac{\partial H}{\partial y}\)
Maximize \(\int ^T _0 -(1 + u^2)^{1/2}\)subject to \(y' = u\)and \(y(0) = A\) \(y(T) =\) free
\(H = -(1 + u^2)^{1/2} + \lambda u\)
\(\frac{\partial H}{\partial u} = -\frac{1}{2}(1 + u^2)^{-1/2}(2u) + \lambda = 0\)or, \(u(t) = \lambda(1 - \lambda ^2) ^{-1/2}\)
\(\lambda ' = -\frac{\partial H}{\partial y} = 0\)
\(H(t, y, u*, \lambda) \geq H(t, y, u, \lambda)\)\(y' = \frac{\partial H}{\partial \lambda}\)\(\lambda ' = -\frac{\partial H}{\partial y}\)
\(\lambda(T) \geq 0 ; y_T \geq y_{min} ; (y_T - y_{min}) \lambda (T) = 0\)
\(\int^T _0 G(y, u) e^{-rt} dt\)
\(H_c \equiv He^{rt} = G(y, u) + \mu f(y, u)\)\(\mu \equiv \lambda e^{rt}\)
Maximize \(\int ^T _0 U(C(t))e^{\delta t} dt\)subject to \(K' = rK(t) - C(t)\)and \(K(0) = K_0 ; K(T) \geq 0\)
\(H = U(C(t)) e^{- \delta t} + \lambda (t) [rK(t) - C(t)]\)
\(\frac{\partial H}{\partial C} = U' (C) e^{- \delta t} - \lambda = 0\)\(K' = rK(t) - C(t)\)\(\lambda ' = -\frac{\partial H}{\partial K} = -r \lambda\)
CARACTERÍSTICAS QUE DEBE DE TENER UN INVESTIGADOR