NR-714 · Week 4 of 8 · Data extraction and the evidence table

NR-714 Week 4 Data Extraction and the Evidence Table: How to Write It

The short answer

Extraction is where a set of papers becomes a data set. The stage asks you to pull the same fields from every included study into a structured form, appraise each study's risk of bias with a tool matched to its design, and present the result as an evidence table a reader can scan across rather than a series of summaries read down. The hard part is comparability: outcomes with the same name are frequently defined differently, and reconciling those definitions in writing is the analytic work. Your section may print this as NR 714 or NR714; it is the same course. Chamberlain publishes no syllabi outside Canvas. The placement here is our teaching judgment from the course's catalog arc; your section's rubric decides what your week actually asks.

NR-714 Week 4 grading scale at Chamberlain, the criterion levels this assessment is scored on, from Chamberlain Tutors
How Chamberlain grades NR-714 Week 4, visualized by Chamberlain Tutors.

What NR-714 Week 4 asks for

A review of transitional care models turns up nine eligible studies, all reporting readmission. One counts readmissions within thirty days to the index hospital only. Another counts any acute admission within ninety days including admissions from the skilled nursing facility to a different hospital. A third reports the proportion of patients readmitted at least once, which is a different quantity from the number of readmissions per patient. A fourth counts emergency department visits alongside admissions in a combined utilization measure. Four papers, one word, four numbers that cannot sit in the same column without a decision. Extraction is where that decision gets made and recorded.

The extraction form is the instrument, and it should be built from the protocol's data item list before the first paper is opened. Typical fields fall into five groups: citation and study identity, including funding and conflicts; design and setting, including country, care setting and dates of data collection; participants, including eligibility, numbers randomized or enrolled, key characteristics and attrition; intervention and comparator, described by components rather than by name; and outcomes, with definitions, measurement instruments, timing, and the effect data in the form actually reported. For this course, add the economic fields, because cost data extracted later means reopening every paper.

Risk of bias assessment runs alongside extraction and must be matched to design. Trials, non-randomized comparative studies, before-and-after improvement reports and qualitative work each have their own instruments and their own domains, and applying a trial-oriented tool to a single-site improvement report produces a judgment that is technically severe and practically uninformative. Record a judgment per domain with a supporting quotation or a page reference, not a global score, because the domain pattern is what feeds the certainty rating in the final stage.

The evidence table is the deliverable readers actually use. It reads across a row for one study and down a column to compare studies, which means every cell in a column must contain the same kind of thing in the same units. Expect a written extraction and appraisal report with a full evidence table, often an appendix of risk of bias judgments, and sometimes a discussion post on an appraisal difficulty. Posts do not reopen once submitted in Canvas, so quote the paper rather than characterizing it.

The NR-714 Week 4 method, step by step

Six moves from nine papers to a table a synthesis can be built on.

  1. Construction and piloting of the extraction form

    Build every field from the protocol, then pilot on two studies of different designs and revise before extracting the rest. Fields discovered late cost a full re-read of everything already done.

  2. Decomposition of interventions into components

    Record what was actually delivered, by whom, how often, for how long and in what setting, rather than the label the authors used. Two interventions sharing a name and sharing no components should not be pooled.

  3. Verification of outcome definitions before recording numbers

    For each outcome capture the exact definition, the window, the instrument and the denominator. Where definitions differ across studies, note the difference in the table rather than flattening it into a single column heading.

  4. Capture of effect data in the form reported

    Record counts and denominators for binary outcomes, means with a measure of spread and the sample size for continuous ones, and note whether the reported estimate was adjusted and for what. Do not convert or calculate at this point.

  5. Appraisal of risk of bias by domain with evidence

    Use an instrument matched to design, judge each domain separately and attach a quotation or a page reference to every judgment. A domain rated without support is an opinion your reader cannot check.

  6. Assembly of the evidence table for comparison down columns

    Order studies deliberately, keep units consistent within each column, and mark missing data as not reported rather than leaving a blank cell that could mean absent or overlooked.

A layout and word budget for an extraction and appraisal report

Our frame for the extraction deliverable, sized for roughly 1,400 to 1,800 words plus the evidence table and appraisal appendix. It is our own outline rather than anything the university issues, and your week's rubric outranks it wherever the two disagree.

SectionWhat belongs in itWord target
Extraction processThe form's origin, the piloting done, who extracted, how accuracy was checked and how author queries were handled.200 to 260
Characteristics of the included setDesigns, countries, settings, sample sizes and dates, written as a paragraph a reader can hold in mind.240 to 300
Intervention descriptionComponents, intensity, duration and delivery personnel across studies, with the variation made explicit.260 to 320
Outcome definitions comparedHow each study defined the primary outcome, the windows used, and the decision you made about comparability.260 to 320
Risk of bias findingsThe tools used by design, the domain pattern across studies, and the studies driving concern with reasons.260 to 320
Readiness for synthesisWhat the table shows about whether pooling is defensible and where narrative synthesis is the honest choice.200 to 260

Evidence craft for extraction writing

Extract what is reported, not what you infer. If a paper does not state an attrition figure, record not reported rather than deriving one from a table. Inference belongs in the synthesis, clearly labeled, and never inside a data field.

Preserve original units and add conversions in a separate column. Keeping the reported figure visible lets a reader check your conversion. A table showing only converted values asks for trust that a methods paper should not require.

Quote the paper when you judge it. A risk of bias domain rated with a short quotation and a page number is verifiable. The same rating with a paraphrase is a claim about a document the reader has to reconstruct.

Record funding and conflicts as extracted fields. They are part of the appraisal picture, they are frequently reported in small print at the end, and a reviewer who notices you captured them reads the extraction as thorough.

Say what you did about missing data. Whether you contacted authors, whether they replied, and what you did in the absence of a reply all belong in the methods. Silence about missing data is read as inattention rather than as absence.

Five mistakes that cost points in this week's territory

  • An evidence table of paragraphs. Cells containing several sentences of summary cannot be compared down a column, which is the only reason the table exists.
  • Outcomes pooled by name. Thirty-day readmission to one hospital and ninety-day utilization across a region are different measures, and merging them silently invalidates everything downstream.
  • One appraisal tool for every design. A trial instrument applied to a before-and-after improvement report yields uninformative judgments and signals unfamiliarity with the toolset.
  • Global quality scores. A single number collapsing several domains hides which weakness matters and is out of step with current appraisal practice.
  • Blank cells. A blank could mean not reported, not applicable or not checked, and the ambiguity undermines the reader's confidence in the whole table.

Before you submit

  • The extraction form was piloted and the piloting is described
  • Interventions are recorded by component, not by label
  • Each outcome carries its definition, window, instrument and denominator
  • Effect data appears as reported, with any conversion shown separately
  • Risk of bias is judged by domain with a quotation or page reference
  • Missing data is marked explicitly and the handling described
  • Every reference appears in the text and every in-text citation appears in the list

Building the NR-714 evidence table?

Send the rubric and your included studies out of Canvas. A premium original draft comes back in 24 to 48 hours with comparable columns, component-level intervention description and domain-level appraisal, and revisions run until the grade lands.

Questions students ask about this stage

How many studies is a reasonable number to include?
There is no target, and chasing one distorts the review. What the number has to be is the honest product of your criteria applied to your search, which for a tightly specified question in a post-acute setting is often between five and fifteen. A small set is not a failure; it is a finding about the evidence base, and it changes what your recommendation can claim. If your set is very large, the usual causes are an under-specified population or an intervention defined by label rather than by component, and the fix belongs in the protocol as a recorded amendment rather than in quiet exclusions. If it is very small, report the near misses and consider whether a defensible widening of the setting or the comparator would capture evidence that genuinely transfers. Either way, the number is reported and explained, never engineered.
Should I contact authors for missing data?
It is standard practice in full systematic reviews and often impractical inside an eight-week session, so the workable position is to try where the missing item is central and to report exactly what you did. If the primary outcome is reported without a measure of spread, or a subgroup you need is described but not tabulated, one short polite message to the corresponding author is worth sending, and the attempt is reportable whether or not it succeeds. Give a deadline for yourself, note in the methods that authors were contacted and how many responded, and state what you did in the absence of a reply, which is usually to exclude that study from a pooled estimate while retaining it in the narrative synthesis. What you never do is estimate the missing number and present it as extracted.
My studies use different instruments for the same construct. What now?
Record each instrument by name with its scale range, direction and any interpretation anchors, then decide whether the construct is close enough to combine. Sometimes it clearly is, in which case a standardized measure of effect allows pooling across scales and you say so. Sometimes it clearly is not, because two instruments named for the same construct measure different aspects of it, and then the honest route is narrative synthesis grouped by instrument. The decision belongs in the text with its reasoning, and it is one of the places where a doctoral reader is looking for judgment rather than procedure. This question also sets up the outcome measurement stage later in the session, where instrument selection and its psychometric evidence become the subject in their own right.

Keep going

Online now