PICO Questions

A focused question should be relevant to a particular clinical scenario and articulated in a manner that can inform an appropriate search strategy. One approach, often utilized when performing a systematic review, is to parse a question into four different parts:

• The patient, population, or health problem being addressed.
• The intervention or exposure being considered.
• The comparison intervention or exposure.
• The clinical outcome(s) of interest.

Framing a question in this manner is often abbreviated as a “PICO” question, where the “P” represents a patient, a population, or a health condition; “I” represents an intervention or an indicator; “C” represents a comparison or control; and “O” represents an outcome. Sometimes a “T” is added—PICO(T)—representing a time element or even a type of study.

A simple example that illustrates a PICO(T) question is assessing the outcome of treating patients with aphthous ulcers with systemic glucocorticosteroids (GCS). Patients with major aphthous ulcers (Patient) that are treated with systemic GCS (Intervention) are compared to patients not receiving systemic GCS (Comparison) to assess the healing of the aphthous ulcers (Outcome) during a time period ranging from 10 to 14 days (Time). This question can be made even more granular by, for example, including only male patients, who receive only a specific dose of systemic GCS, and comparing these patients with patients who receive oral mouth rinses containing GCS (Table 29‐1).

PICO and similar frameworks need to be tweaked according to the type of question being asked. For example, clinical questions concerning therapy, prevention, diagnosis, prognosis, or etiology may have to be framed differently (Table 29‐2). Utilizing a PICO(T) framework is also helpful when treating patients in order to formulate differential diagnoses.

PICO questions are often referred to as foreground questions, as these are more specific than background questions that address more general knowledge. Differentiating between foreground and background questions facilitates formulating a search strategy.

Search Strategies

Begin with a broad search, using only a few terms, and do not apply any search filters. There are several databases and resources available that are populated with clinical practice guidelines, synopses, systematic reviews, primary studies, and other information (Table 29‐3). (Tip: Use nouns as search terms rather than adjectives to retrieve literature that contains quantifiable results.) If researching a clinical question, the PICO framework (see above) components of population and intervention are useful in determining initial search terms. To narrow results, add additional terms using Boolean logic (Figure 29‐1). Next, employ search filters to limit results to relevant material—useful search filters include limits such as language, publication date, and study design type. If the search is retrieving too many results or too many irrelevant results, narrow the search by searching a specific field (e.g., title or abstract). Finally, if the results of the initial search are not relevant, try using Boolean logic to include synonyms (e.g., “pregnancy” for “pregnant women”) and alternative spellings (e.g., UK and US spellings) for search terms.

Many medical databases use controlled vocabulary—the preferred terminology used within a particular database to describe and index materials. When conducting extensive searching, examine the databases’ thesauri to identify synonyms for search terms. (Note: most medical databases will correct common misspellings, recognize abbreviations, and automatically search plurals.)

Databases and Other Resources

In order to conduct a thorough search for relevant information, search a combination of databases. Although the determination of which databases to search depends on the topic/question at hand, at a minimum search Cochrane (the gold standard for evidence‐based information), Trip (the best “preappraised” resource), and PubMed (the most comprehensive medical literature search engine; see below for database descriptions and individual search tips). Search all three of these major databases in order to avoid missing relevant material; there is less overlap among these resources than one would assume, because they cover different types of materials and the material in each is updated on different cycles. Each of these databases has a unique underlying architecture, vocabulary, and results algorithm, and therefore will return different results.

A search strategy for evidence in order to decide a course of action within a clinical setting or answer a clinical question will differ from a search strategy for conducting a literature review or a systematic review. The latter are comprehensive, the former is not. If a search is being conducted to answer a clinical question, include high levels of evidence (e.g., guidelines or systematic reviews).

If an initial search results in too many irrelevant items, the following strategies may be useful:

• Limit the search to the title and abstract.
• Limit the results to secondary, preappraised literature such as clinical practice guidelines, summaries, critical summaries, and systematic reviews.
• If the search does not result in secondary literature, one strategy to narrow down the result of primary studies is to limit the search to randomized controlled trials (RCTs).
• If a search retrieves too many irrelevant results, searching using subject headings (rather than keywords) might prove to be a more efficient search strategy. Searching by subject headings will narrow the result of the search to only those items for which the subject heading in question has been considered (indexed) as the main topic of the item. Most of the databases listed here include specialized thesauri that list the subject headings used to index articles contained in the databases.
• At least preliminarily, set aside results published a decade or more ago.
• Add additional PICO components to your search terms, beginning with the outcome category.

Table 29‐4 contains a list of databases that contain oral medicine–related information. An effective search strategy would begin with a search in Cochrane, followed by a search of Trip (particularly if Cochrane does not return relevant results), and then a search of PubMed to cover any potential gaps in results. Cochrane contains materials with high levels of evidence, such as systematic reviews, and is therefore an effective starting place. Because Cochrane only includes secondary materials, however, it is a smaller database than many of the others listed, and therefore may not contain all the information relevant to the topic being searched (including the most recent information). PubMed, by contrast, is the largest database listed, and will always retrieve the most results.

Begin with a search using a combination of keywords and subject headings in order to gather a sense of the results and to capture results that might be so new that they have not been assigned subject headings at the time of the search. Cochrane and PubMed share a thesauri of index terms—both databases use MeSH (Medical Subject Headings) for subject headings (see below for a detailed explanation of MeSH). Trip does not use a specialized thesaurus. After conducting the initial search, note any useful related terms and synonyms, and revise the search using all relevant search terms.

Type of Clinical Information

Clinicians informing their practice using scientific literature are interested in answering questions directly related to their day‐to‐day activities. As mentioned before, such questions are classified according to their nature in four clinical areas:^1,2

• Diagnosis: establishing the ability of a test to discriminate patients with a disease or conditions from those without the disorder.
• Therapy or prevention: determining the effect (desirable and undesirable) of an intervention on relevant outcomes (e.g., patient‐centered outcomes).
• Harm/etiology: determining the effects of an intervention or exposure considered potentially harmful on outcomes relevant to patients (e.g., downsides or adverse effects).
• Prognosis: predicting the future course of a patient’s recovery or aggravation from a disease.

It is recommended that users of the literature correctly determine the type of question to address, as achieving this step will allow the question to be linked to the most appropriate study design (Table 29‐5). When addressing therapeutic questions, experimental designs using chance (analogous to tossing a coin) rather than convenience to allocate participants to a treatment of interest that is compared with another active intervention, placebo, or no treatment are preferable (RCTs). In the absence of such experimental types of study, clinicians can also use nonexperimental or observational types of design. In this type of study, not chance but clinicians, patients and their preferences, or simply the circumstances determine the allocation of participants to treatment arms or an exposure.

In theory, one can also answer harm questions using RCTs; however, when a potential hazard is apparent (e.g., a serious, initially unintended adverse event), the intentional assignment of participants to the potential harmful intervention is not ethical. In this case, observational studies (cohort and case–control studies) are helpful to learn about the eventual harmful effect of an exposure or intervention that has occurred or will occur, without the need for the investigator to assign participants to the study arms.

As no diagnostic test is perfect, clinicians who want to learn about the performance or accuracy of diagnostic strategies for replacing a preexisting test because it is cheaper or less invasive, for triaging subsequent test applications (e.g., a positive result warrants further investigation to confirm diagnosis, while a negative result rules out the disease), or for adding on a new test to a preexisting diagnostic workout will choose a particular type of study design. Diagnostic test accuracy studies recruit participants with a suspicion of having the condition of interest and undergo an index test or new test, while also receiving a reference test or gold standard. Then, investigators report how often the index test is in agreement and disagreement with the reference test.³

Questions of prognosis examine the presence of factors that can positively or negatively influence the future course of action of a disease or condition, with the purpose of increasing the chance of achieving better outcomes. These studies include patients with and without a particular feature or prognostic factor (e.g., age, comorbidity, educational status, etc.) and examine the extent to which this factor influences a particular outcome in the future.

It can be inferred from the variety of clinical questions and study design indications presented here that the idea of a single “evidence pyramid” placing RCTs at the top of the hierarchy as the most suitable study to inform all clinical decisions is an oversimplification, as such a depiction ignores the idea that study design on its own is not a fair indicator of trustworthiness.⁴ The traditional evidence pyramid is more appropriate for being applied to questions about therapy. Thus, it is strongly recommended to abandon the use of this evidence pyramid and replace it with a broader perspective, in which not a single primary study but rather a body of evidence (i.e., the group of studies informing the effect of an intervention of exposure on a particular outcome) is considered for decision‐making. The certainty that clinicians can place in this evidence then not only focuses on the risk of bias of the studies at hand, but also the consistency across their findings (i.e., studies showing similar results), the precision of their estimates (i.e., width of confidence intervals), the relevance or directness of the study to the question at hand (i.e., appropriate generalizability or applicability), and the possibility for publication bias (i.e., the extent to which all studies conducted are made available irrespective of their results).^1,5,6 (For more details on study design, see Chapter 1: Overview of Clinical Research).

It is not uncommon that a new and somewhat controversial primary study attracts clinicians’ and patients’ attention, including presentations of the findings in media channels. However, clinicians should favor systematic reviews and evidence‐based clinical practice guidelines over a single study, as systematic reviews and evidence‐based clinical practice guidelines provide the full picture of all available evidence regarding a clinical question, making them more informative for the busy practitioner. In the same way that “one swallow doesn’t make a summer,” a single study outcome must be put in the context of the totality of the evidence before it is decided that a change in practice is warranted.

Causation, Association, and Correlation

Causation is important for clinical practice. Any practitioner trying to determine what is the best therapeutic strategy to treat a condition or disease, or looking for risk factors and studying their influence in the occurrence of a future event, is framing a causal question. Despite seeming simple, causal inferences are complex. In a hypothetical world where anything is possible, an investigator interested in studying the effect of the human papilloma virus (HPV) vaccine on the incidence of oropharyngeal squamous cell carcinoma (OSCC) will administer the vaccine to a group of participants, preferably a few thousand of them, and then perform follow‐up assessments to determine how many did and did not develop OSCC. In this hypothetical world, the investigators would now time‐travel back to the time before the vaccine was initially administered and enroll the same people who, in the first scenario, were vaccinated against the HPV virus and this time not administer the vaccine. The investigators would again perform follow‐up assessments and determine how many did and did not develop OSCC. Thus, in these hypothetical scenarios, the same people were both vaccinated and not vaccinated and were essentially acting as their own control group. However, in the real world where one only has access to one side of the story at a time and the alternative scenario is missing and unobserved, this is called the “counterfactual dilemma.”⁷

Since time‐traveling remains possible only in theory, researchers will try all sorts of methodologic strategies to access that inaccessible alternative scenario. One strategy is to have a control group that could stand in for the second (unobserved) scenario. As inferred from the time‐traveling idea, it is necessary to have a second observation—a control group—that is as similar as possible to the group receiving the experimental intervention. Hence, any time that two groups under comparison differ, the trustworthiness of an association or causality claim weakens. Additional information on causality and its frameworks can be found elsewhere.⁸

The terms “causation” and “association” are often used interchangeably, but while causation implies association, the opposite is not true. For example, people who consume four or more cups of coffee a day have a lower risk of presenting with actinic cheilitis (AC). This does not mean that consuming coffee provides a protective effect from suffering from AC; rather, an alternative explanation would be that those consuming a larger amount of coffee a day usually work in indoor environments, such as an office, which dramatically reduces their exposure to direct sunlight, a known risk for AC. In this scenario, sunlight exposure is a confounder—that is, an alternative explanation for both the exposure (sunlight) and the outcome (AC)—for the association between coffee consumption and AC. Thus, a causal connection cannot be claimed, and the hypothesized association is spurious.

Nor can the terms “association” and “correlation” be used interchangeably. An association refers to one variable or clinical feature providing information about another variable or clinical feature. A correlation refers to one variable or clinical feature increasing or decreasing a fixed amount for a unit of increase or decrease of the other variable or clinical feature (Figure 29.2). Thus, “correlation implies association, but not causation. Conversely, causation implies association, but not correlation.”⁹