NR-701 · Week 2 of 8 · Risk of bias in quantitative studies

NR-701 Week 2 Risk of Bias in Quantitative Studies: How to Write It

The short answer

Once the search is defensible, the appraisal starts, and the doctoral version of appraisal is domain-based rather than impressionistic. You judge a quantitative study one bias mechanism at a time: how people got into groups, whether anyone knew which group they were in, whether the outcome was measured the same way for everyone, what happened to the people who disappeared, and whether the reported results are the results that were planned. Then you say which direction each problem pushes the finding. Your section may print this as NR 701 or NR701; 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-701 Week 2 grading scale at Chamberlain, the criterion levels this assessment is scored on, from Chamberlain Tutors
How Chamberlain grades NR-701 Week 2, visualized by Chamberlain Tutors.

What NR-701 Week 2 asks for

A telehealth service considering remote monitoring for chronic obstructive pulmonary disease finds a study reporting fewer admissions among monitored patients. The abstract is clean. The methods section says patients were offered monitoring and those who accepted were compared with those who declined. That single sentence changes everything the study can support, because the people who accept a monitoring program differ from the people who decline it in ways that also predict admission. Nothing was randomized, so the comparison is between two populations rather than between two treatments, and the effect reported belongs partly to the difference between them.

That is what this stage teaches: reading for the mechanism by which a result could be wrong. Selection bias arises when the route into a group is related to the outcome. Performance bias arises when groups get different care beyond the intervention itself, which is why unblinded studies of behavior-heavy interventions carry it almost automatically. Detection bias arises when outcome measurement differs between groups, and it is worst when the outcome is a judgment made by someone who knows the assignment. Attrition bias arises when loss to follow-up differs in size or in kind between groups. Reporting bias arises when the outcomes published are the outcomes that turned out well.

Design tells you which mechanisms are plausible before you read a line of the results. Randomized designs address selection at the point of allocation and are still vulnerable everywhere else. Prospective cohorts handle sequence and remain exposed to confounding by indication. Case control work is efficient and lives or dies on control selection. Interrupted time series and controlled before-and-after designs are common in the practice literature a doctoral student needs, and they carry the specific vulnerability that everything else in a health system also changes over time.

For a practice doctorate this is not academic. You are deciding whether to build a change at a site on this evidence, and the question is not whether the study is flawed, because all of them are, but whether the flaws could plausibly account for the effect you would be counting on. Expect a written appraisal of one or more studies, often with a domain table, and possibly a board post naming the single most consequential threat you found. Post that judgment rather than a summary: posts do not reopen once submitted in Canvas.

The NR-701 Week 2 method, step by step

Six analytic moves that produce an appraisal a committee could act on.

  1. Reconstruct the design from the methods in your own words

    Write one sentence describing what was actually done to whom and when, without using the label the authors chose. If your sentence does not match their label, you have found something worth writing about.

  2. Establish how people entered their groups

    Randomized, self-selected, assigned by clinician judgment, or defined retrospectively from records. This single fact governs how much of the observed difference can be attributed to the intervention at all.

  3. Locate who knew what, and when

    Blinding of participants, of those delivering care, and of those measuring the outcome are three separate questions with three separate consequences. Where blinding was impossible, say what was done instead, such as an objective endpoint or an independent assessor.

  4. Follow the missing people

    Compare numbers enrolled, numbers completing and numbers analyzed. Then ask whether the losses differ between groups and whether the reason for leaving could be related to the outcome, which is when attrition becomes bias rather than noise.

  5. Check the outcomes reported against the outcomes planned

    Look for a registration record, a stated primary outcome, or a protocol reference. A results section that emphasizes a secondary endpoint while the primary one is mentioned briefly is a pattern worth naming carefully and without accusation.

  6. Assign a direction to each threat and then a verdict

    For each domain say whether the problem would inflate, deflate or unpredictably distort the effect, then state how much weight the study should carry in a decision about your setting. Direction is what separates appraisal from a list of complaints.

A layout and word budget for a risk of bias appraisal

Our frame for appraising one study in depth, sized for roughly 1,500 to 1,900 words. 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
Decision contextThe practice decision this study is being read for, so the appraisal has a purpose a reader can weigh it against.140 to 180
Design reconstructedWhat was done to whom and when, written from the methods without using the authors' label.200 to 250
Group formationAllocation or entry route, concealment where relevant, and baseline comparability with the numbers that show it.260 to 320
Delivery and measurementWho knew the assignment, how the outcome was captured, by whom, and with what instrument or definition.280 to 340
Missing dataLosses by group, stated reasons, the analysis population used, and how the authors handled the gaps.220 to 280
Domain judgments with directionEach bias domain rated with a reason and the direction it would push the estimate.280 to 340
Weight for the decisionHow much this study should count toward the change you are considering, and what would change that.160 to 200

Evidence craft for bias appraisal

Use a published appraisal instrument matched to the design. Tools for trials, for non-randomized studies of interventions and for observational work ask different questions, and naming the instrument lets a grader check your reasoning against a standard rather than against your intuition.

Quote the sentence you are judging. When you conclude that allocation was not concealed, point to the words in the methods that told you so, or note that the paper is silent. Unsupported bias judgments are the fastest way to lose a doctoral appraisal row.

Distinguish not reported from not done. A paper that does not describe blinding may still have blinded. Write the uncertainty accurately, rate the domain as unclear where that is the honest rating, and say what information would resolve it.

Give counts and denominators for every group figure. Sixteen of 84 lost from the monitored arm compared with five of 81 from usual care is an attrition finding. A statement that dropout was higher in the intervention group is an impression.

Keep causal verbs matched to the design. Reduced belongs to assigned comparisons. Was associated with belongs to observational ones. In an appraisal course this distinction is graded directly and repeatedly.

Five mistakes that cost points in this week's territory

  • Generic limitations. Small sample, single site and further research is needed describe most studies ever published and demonstrate nothing about this one.
  • Bias named without a mechanism. Saying a study has selection bias is a label. Saying that patients who accepted monitoring were healthier at entry, which would inflate the apparent benefit, is appraisal.
  • Ignoring the analysis population. A study that analyzes only completers has answered a different question from one that analyzes everyone allocated, and the difference frequently matters more than the design label.
  • Treating any flaw as fatal. Doctoral appraisal grades studies rather than dismissing them, and a paper that rejects everything has produced no usable evidence for a decision.
  • No decision at the end. The graded task is to say what this study is worth for a practice change, and an appraisal without that sentence has stopped one step short.

Before you submit

  • The practice decision the appraisal serves is stated up front
  • The design is reconstructed from the methods rather than taken from the title
  • The route into each group is named explicitly
  • Blinding is addressed separately for delivery and for measurement
  • Attrition appears as counts by group with reasons
  • Each domain judgment carries a direction of likely bias
  • The appraisal closes with the weight the study earns and why

Appraising studies for NR-701?

Send the rubric and the article out of Canvas. A premium original draft comes back in 24 to 48 hours with each bias domain judged from the methods and given a direction, and revisions run until the grade lands.

Questions students ask about this stage

The paper never says whether outcome assessors were blinded. What do I write?
Write that the paper does not report it, rate the domain as unclear, and then reason about how much it would matter here. The consequence depends on the outcome: a laboratory value or an administrative admission count is relatively resistant to assessor knowledge, while a symptom score, a clinical judgment of severity or an adjudicated event is not. So the same silence is a minor issue in one study and a serious one in another, and saying which case you are in demonstrates exactly the judgment the domain approach exists to produce. Avoid the two easy errors: assuming blinding happened because a study is otherwise well conducted, and assuming it did not because reporting is incomplete.
How should I treat a single-site improvement report in an appraisal?
As real evidence about feasibility and as weak evidence about effect. These reports are abundant in the practice literature and they carry information trials rarely do: how the change was staffed, what resistance appeared, what had to be adjusted in the second month. What a before-and-after account at one site cannot do is separate the intervention from everything else that shifted during the same period, including seasonal variation, concurrent initiatives and the attention the project itself created. Appraise it on its own terms, name the rival explanations, and use it for the implementation layer of your reasoning while letting stronger designs carry the claim about whether the intervention works.
Is it acceptable to appraise a study whose statistics I cannot follow?
Yes, provided you are precise about what you can and cannot judge. Most bias domains are readable from the methods without any statistical training: who entered which group, who knew what, how the outcome was captured, who went missing. Handle the analysis honestly by naming the technique accurately, stating in one sentence what class of question it addresses, and reporting the result in the units the authors used. Then say plainly that the appropriateness of the modelling is outside what your appraisal assessed. That sentence costs nothing and protects everything else in the paper, whereas statistical vocabulary used decoratively is easy for a doctoral grader to spot and expensive when it is spotted.

Keep going

Online now