# Microbiology Practice Questions: Template and Examples A practice question is a diagnostic test you run on a learner, not on a patient. Built well, it shows exactly which concept failed and why. Built badly, it rewards word matching and tells you nothing. By the end of this guide you will be able to draft a four-option microbiology item with a written rationale for every option, map that item to a cognitive level, blueprint a full set across a course, run the item statistics that reveal broken questions, and rewrite the ones that fail. You will also have six complete worked examples and a machine-checkable template you can reuse for any microbiology course. What to have on hand: a plain-text editor, Python 3 (standard library only, no packages needed), a spreadsheet for the blueprint, and one primary reference per item so the answer key can be defended. A calculator is enough for the statistics, though a spreadsheet is faster. ## What a Microbiology Practice Question Set Is For Microbiology practice questions do three jobs. They force retrieval, they expose misconceptions, and they give the instructor a measurement of where a cohort actually stands. All three depend on item quality. The evidence on item quality is sobering. A global panel of 22 pharmacology educators evaluated roughly 200 educator-written multiple-choice questions against four separate frameworks: item writing guidelines, Bloom's taxonomy, a core concepts map, and item design format. Nearly 80% of items addressed at least one core concept and 52% used a context-dependent format. Only 40% reached higher levels of Bloom's cognitive domain, and just 10% adhered to all of the item writing guidelines [1]. Those are expert-written items from a funded initiative. Unreviewed question banks do worse. A strong positive correlation was observed between the context-based item format and item effectiveness in that same analysis [1]. Context-based means the stem presents a short clinical or laboratory scenario and the learner must apply a principle to it. That format is the default for a serious microbiology practice set. Practice sets also work as pre and post measures. A prospective before-and-after study of a [clinical microbiology](/knowledge/diagnostics/microbiology/clinical-microbiology-from-specimen-collection-to-pathogen-identification) bioinformatics escape game used two distinct 10-item tests covering the same predefined learning objectives, administered before and immediately after the activity, with 55 participants split between second-year medical students and postgraduate MDs and PharmDs working in teams of three to four [2]. The design point matters more than the game: two parallel forms, same objectives, tested twice. That is the minimum credible structure for any practice set you want to use as an assessment. ## The Microbiology Practice Question Template: Eight Fields Every item in a defensible bank carries eight fields. Leave one out and the item becomes hard to audit, hard to revise, and hard to retire when the underlying evidence changes. | Field | What goes in it | What breaks without it | |--|--|--| | `id` | Stable identifier, for example `BAC-AST-014` | Cannot track an item across revisions or link statistics to content | | `domain` | Content area, for example bacteriology, virology, genomics, infection control | Blueprint coverage cannot be measured | | `bloom` | Cognitive level, 1 to 6 | The set drifts toward recall without anyone noticing | | `stem` | The scenario or factual setup | The item has no context to apply knowledge to | | `lead_in` | The actual question, ending in a question mark | Learners answer the wrong question correctly | | `options` | Three to five choices, one key and the rest distractors | Guessing inflates scores | | `rationale` | A written reason for every option, right and wrong | Wrong answers teach nothing | | `source` | The reference that supports the key | The item cannot survive challenge | A distractor is a wrong option designed to attract learners who hold a specific misconception. A key is the single correct option. The stem is the scenario. The lead-in is the question itself. Keeping stem and lead-in as separate fields is the single most useful structural choice in the template, because most flawed items fail at the join between them. ### The stem Write the stem as a compact laboratory or clinical scene. Give the specimen, the method, and the finding. Two or three sentences is enough. Avoid negative stems ("which of the following is NOT") unless the negative is the actual skill being tested, such as recognizing an organism that does not grow on MacConkey agar. If you must use a negative, set it in capitals and bold it in the rendered text. Do not put the answer inside the stem. A stem that says "cefiderocol, a siderophore cephalosporin that enters through iron transport" and then asks how cefiderocol enters has answered itself. ### The lead-in The lead-in is the question. It ends with a question mark. It should be answerable without reading the options, which is the classic cover-the-options test. Weak lead-in: "Which of the following is true?" Strong lead-in: "Which statement best explains the discordance between the two methods?" A lead-in that begins "Which of the following statements is correct" invites the learner to evaluate four unrelated claims. That is four true/false items wearing a trench coat, and it measures almost nothing. ### The options Use four options. Three increases guessing to 33%. Five adds writing burden without much gain for most microbiology content. Every distractor should represent a named misconception. Before you write the option, write the misconception in a note. If you cannot name the misconception, the distractor is filler and will sit unselected. Keep options parallel in grammar and length. A correct option that is noticeably longer and more qualified than the others is a cue, and learners will find it. Keep the key's length within a few words of the median distractor. Avoid "all of the above" and "none of the above". Both change the task from content knowledge to test-taking strategy. ### The key and the rationales The key is the letter, not the text, so the key survives option shuffling. Every option gets a rationale, including the key. The key rationale explains why the option is correct and cites the source. Distractor rationales explain what the learner was probably thinking and why that reasoning fails. This is the highest-value part of the template and the one most often skipped. A bank with full rationales doubles as a study resource. A bank without them is a score generator. ### The source anchor Each item needs one primary anchor. That can be a journal article, a standards document, a textbook chapter, or a validated manufacturer package insert. For antimicrobial susceptibility items, anchor to the breakpoint table version in force at the laboratory, because breakpoints change and an item tied to a superseded table becomes wrong without anyone editing it. ## A Worked Template You Can Copy and Check Here is a seven-step walkthrough from a blank file to a validated bank. 1. Create `questions.json` and store items as a list of objects using the eight fields. 2. Write the item, then immediately write the misconception note for each distractor. 3. Assign the Bloom level using the table in the next section. 4. Attach the source anchor with enough detail that a colleague could find the exact passage. 5. Run the structure checker below. 6. Fix every reported problem. 7. Freeze the item and schedule a review date. A single item in the file looks like this: ```json [ { "id": "BAC-AST-014", "domain": "bacteriology", "bloom": 4, "stem": "A laboratory reports cefiderocol as resistant for a carbapenem-resistant Acinetobacter baumannii isolate by disk diffusion. The same isolate is susceptible by broth microdilution performed in iron-depleted medium.", "lead_in": "Which statement best explains the discordance?", "options": [ "Broth microdilution in iron-depleted medium is the standardized method, and disk diffusion may overcall resistance.", "Disk diffusion is the standardized method, and the broth microdilution run failed.", "Cefiderocol cannot be tested in vitro, so both results are invalid.", "A. baumannii is intrinsically resistant to cefiderocol, so susceptibility is a [transcription error](/knowledge/molecular-biology/transcription-error)." ], "key": 0, "rationale": { "0": "Correct. Breakpoints differ across standards organizations, broth microdilution requires iron-depleted media, and disk diffusion is practical but may overcall resistance.", "1": "Reverses the method hierarchy. Disk diffusion is the practical but less reliable option here.", "2": "Cefiderocol can be tested, but the method is demanding and requires iron-depleted medium.", "3": "Cefiderocol retains activity against carbapenem-resistant A. baumannii and is reserved for it when preferred alternatives are unsuitable." }, "explanation": "Discordant cefiderocol results are a recognized testing problem tied to iron availability and method choice.", "source": "Cefiderocol review, Curr Opin Infect Dis" } ] ``` Save this checker as `check_bank.py`: ```python import json, sys REQUIRED = ["id", "domain", "bloom", "stem", "lead_in", "options", "key", "rationale", "explanation", "source"] def audit(items): problems = [] for it in items: missing = [f for f in REQUIRED if f not in it] if missing: problems.append((it.get("id", "?"), "missing fields: " + ", ".join(missing))) continue if not isinstance(it["key"], int) or it["key"] >= len(it["options"]): problems.append((it["id"], "key index out of range")) if len(it["options"]) < 3: problems.append((it["id"], "fewer than 3 options")) if set(map(str, it["rationale"].keys())) != set(map(str, range(len(it["options"])))): problems.append((it["id"], "rationale does not cover every option")) if it["bloom"] not in (1, 2, 3, 4, 5, 6): problems.append((it["id"], "bloom level out of range")) if not it["lead_in"].strip().endswith("?"): problems.append((it["id"], "lead_in does not end in a question mark")) if "NOT" in it["stem"] or "EXCEPT" in it["stem"]: problems.append((it["id"], "negative stem, justify or rewrite")) return problems items = json.load(open(sys.argv[2], encoding="utf-8")) problems = audit(items) print("bank:", sys.argv[2]) print("items parsed:", len(items)) print("problems:", len(problems)) for pid, msg in problems: print(" " + pid + ": " + msg) print("bloom levels used:", sorted({it.get("bloom") for it in items if isinstance(it.get("bloom"), int)})) print("domains:", sorted({it.get("domain", "?") for it in items})) print("status:", "PASS" if not problems else "FAIL") ``` Run it: ``` python3 check_bank.py questions.json ``` Expected output for a clean file: ``` bank: questions.json items parsed: 12 problems: 0 bloom levels used: [1, 2, 3, 4] domains: ['bacteriology', 'genomics', 'infection_control', 'mycology', 'virology'] status: PASS ``` The `domains` line is the part people skip. It tells you at a glance whether the bank covers the course or has quietly collapsed into bacteriology. ## Mapping Items to Bloom's Taxonomy Bloom's taxonomy is a six-level classification of cognitive demand, from simple recall to creating new work. For a microbiology course, levels 1 through 4 cover nearly everything you need. | Level | Verb | Microbiology task | Example prompt | |--|--|--|--| | 1 Remember | define, list, name | Retrieve a fact | "Which medium is required for cefiderocol broth microdilution?" | | 2 Understand | explain, classify, compare | Explain a mechanism or contrast two methods | "Why does iron availability change the result?" | | 3 Apply | calculate, interpret, use | Apply a rule to a new isolate | "Interpret this MIC against the current breakpoint" | | 4 Analyze | differentiate, attribute, explain discordance | Resolve conflicting data | "Why do the two methods disagree?" | | 5 Evaluate | judge, critique, justify | Judge evidence quality | "Should this model's accuracy figure be trusted?" | | 6 Create | design, construct | Design an experiment or a surveillance plan | "Design a sampling scheme for a hospital outbreak" | A course set that is 70% level 1 will produce learners who can recite and cannot interpret. The pharmacology panel found only 40% of expert-written items reached higher Bloom levels [1], which suggests most educator-written banks are skewed low by default. A practical target for a semester-long microbiology course: roughly 20% level 1, 30% level 2, 30% level 3, and 20% level 4. Add a handful of level 5 items for upper-level and graduate courses. Watch for items that are labeled level 3 but are actually level 1. "Which organism is oxidase positive?" is level 1 no matter how it is dressed. "A non fermenting gram-negative rod from a respiratory specimen is oxidase positive and grows at 42 degrees Celsius. What is the most likely genus?" is level 3 because the learner must combine three findings. ## Blueprinting the Set Before You Write a Single Item A blueprint is a grid that fixes how many items come from each content domain and at which cognitive level. Writing items first and blueprinting later always produces gaps. Start with your course objectives. If your microbiology course has eight units, the blueprint has eight domain rows and four Bloom columns. Fill each cell with a target count. | Domain | Level 1 | Level 2 | Level 3 | Level 4 | Total | |--|--|--|--|--|--| | Bacterial cell structure and growth | 2 | 2 | 1 | 1 | 6 | | Culture and identification | 2 | 3 | 3 | 1 | 9 | | Antimicrobial susceptibility testing | 1 | 2 | 3 | 2 | 8 | | Virology and viral diagnostics | 2 | 2 | 2 | 1 | 7 | | Mycology | 1 | 2 | 1 | 0 | 4 | | Genomics and bioinformatics | 1 | 2 | 2 | 1 | 6 | | Infection prevention and control | 1 | 2 | 2 | 1 | 6 | | Host response and immunology | 2 | 2 | 1 | 1 | 6 | | **Total** | **12** | **17** | **15** | **8** | **52** | Fifty-two items supports a semester course with room to retire several after the first statistics run. Retire at least 10% of any new bank per term. Two blueprints are worth maintaining: one for the graded exam and one for the practice bank. The practice bank can be three times larger and include deliberately harder items, because the goal there is learning rather than measurement. ## Microbiology Practice Question Examples Across the Curriculum Each example below is written in the template format. Stem, lead-in, options, key, then the rationale that would appear in the bank. ### Example 1: Entry mechanism (Bloom 1) **Stem:** A clinical isolate of carbapenem-resistant *Acinetobacter baumannii* is submitted for cefiderocol testing. Cefiderocol is a cephalosporin linked to a siderophore. A siderophore is a small molecule that chelates ferric iron and is taken up by bacterial iron transport systems. **Lead-in:** Which entry route accounts for its activity against this organism? - A. Passive diffusion through outer membrane porin channels alone - B. Uptake through bacterial iron transport systems after siderophore conjugation - C. Direct insertion into the outer membrane lipid bilayer - D. Transfer through a type IV secretion system **Key:** B **Rationale:** B is correct. Cefiderocol is the first clinically available antibiotic that exploits bacterial iron transport to enter gram-negative pathogens [3]. A describes the ordinary route for many beta-lactams and misses the defining feature of this drug. C describes a membrane-active agent such as colistin, which is a peptide, not a cephalosporin. D describes a system for transferring DNA and proteins between cells, not a drug uptake route. ### Example 2: Discordant susceptibility results (Bloom 4) **Stem:** A laboratory reports cefiderocol as resistant for a carbapenem-resistant *A. baumannii* isolate by disk diffusion. The same isolate is susceptible by broth microdilution performed in iron-depleted medium. **Lead-in:** Which statement best explains the discordance? - A. Broth microdilution in iron-depleted medium is the standardized method, and disk diffusion may overcall resistance - B. Disk diffusion is the standardized method, and the broth microdilution run failed - C. Cefiderocol cannot be tested in vitro, so both results are invalid - D. *A. baumannii* is intrinsically resistant to cefiderocol, so the susceptible result is a transcription error **Key:** A **Rationale:** A is correct. Breakpoints for cefiderocol differ across standards organizations, broth microdilution requires specialized iron-depleted media, and disk diffusion is practical but may overcall resistance [3]. B reverses the method hierarchy. C is false, since testing is possible but technically demanding. D is false, because cefiderocol retains activity against carbapenem-resistant *A. baumannii* and is reserved for that organism when preferred alternatives are unavailable or unsuitable [3]. This item is a good model for level 4 because it gives the learner two conflicting results and asks for the explanation rather than the answer. ### Example 3: Variant calling from long reads (Bloom 3) **Stem:** A public health laboratory sequences 14 bacterial isolates with Oxford Nanopore Technologies using the super-high-accuracy basecalling model and targets 10x coverage. Variant calling is the process of identifying positions where a genome differs from a reference. **Lead-in:** Which statement about single nucleotide variant detection is best supported by benchmarking data? - A. Only short reads can reach acceptable precision for single nucleotide variants - B. Deep learning variant callers such as Clair3 and DeepVariant reach precision and recall comparable to or better than full-depth short-read data at this depth - C. Long-read data cannot be used for variant calling because of homopolymer errors - D. Variant calling requires duplex reads exclusively **Key:** B **Rationale:** B is correct. In a benchmarking study across 14 bacterial species, deep learning variant callers outperformed traditional methods, and 10x depth of super-high-accuracy long-read data achieved precision and recall comparable to or better than full-depth short-read sequencing [4]. A is contradicted by that finding. C is outdated, because high-performing callers with super-high-accuracy data mitigate homopolymer error [4]. D overstates the requirement, since duplex reads improve accuracy but are not the only acceptable read type. ### Example 4: Infection control competency clusters (Bloom 2) **Stem:** An infection prevention director is building a competency framework for infection control link nurses. A scoping review with cluster analysis identified nine core competency domains and grouped them into three clusters. **Lead-in:** Which reported cluster contains the microbiology and surveillance domain? - A. System-supporting and developmental competencies - B. Infection prevention and control in clinical practice and microbiology and surveillance - C. Leadership and financial management - D. Environmental services and facility engineering **Key:** B **Rationale:** B is correct and matches the cluster named in the review [5]. A is the other reported cluster and does not contain microbiology and surveillance. C and D are plausible-sounding domains that were not among the clusters reported. ### Example 5: Judging a prediction tool (Bloom 5) **Stem:** A machine learning tool reports 96% accuracy for predicting antibacterial peptide activity directly from amino acid sequence. **Lead-in:** What is the most important first question before you trust that figure? - A. Whether the model was trained on a graphical processing unit - B. Whether training and test sets were constructed to avoid data leakage and whether a standard evaluation framework was applied - C. Whether the sequences contain only the 20 standard amino acids - D. Whether the tool returns a result in under one second **Key:** B **Rationale:** B is correct. Existing studies of antimicrobial peptide prediction often rely on heterogeneous data sets and inconsistent evaluation protocols, which can lead to data leakage and inflated performance estimates, and the field lacks a standard evaluation framework [6]. Compute hardware, residue alphabet, and speed do not determine whether a reported accuracy is real. Data leakage means information from the test set has influenced training, so the reported score reflects memorization rather than generalization. ### Example 6: Exposure variability in acute illness (Bloom 3) **Stem:** An acutely ill patient with a gram-negative bloodstream infection receives a beta-lactam dosed by body weight and estimated glomerular filtration rate. **Lead-in:** Why does this approach leave substantial uncertainty about drug exposure? - A. Beta-lactams have no measurable serum concentrations - B. Population-level clearance estimates from weight and estimated glomerular filtration rate produce substantial variability in observed serum concentrations in acutely ill patients - C. Beta-lactams are inactive against gram-negative bacteria - D. Beta-lactam concentrations depend only on oral absorption **Key:** B **Rationale:** B is correct. A one-size-fits-all approach based on weight and estimated glomerular filtration rate results in substantial variability in observed serum concentrations in acutely ill patients, and dose individualization uses measured concentrations from that patient [7]. A is false, since beta-lactam concentrations are measurable and therapeutic drug monitoring is recommended in international guidelines [7]. C is false. D is false, since these agents are typically given intravenously in this setting. ### Short items for rapid review These work as single-line checks in a spaced review deck rather than as exam items. | Stem | Lead-in | Key | Anchor | |--|--|--|--| | Historical presence modeling for *Culex pipiens* | Which two bioclimatic variables were most important? | Altitude and annual precipitation | [8] | | A learner states that all microbes are harmful | Which correction is most accurate? | Many microbes are beneficial and are used in food, medicine, and industry | [9] | | A surveillance model predicts susceptibility with moderate accuracy | What is the appropriate use? | A supportive signal alongside phenotypic testing, not a replacement for it | [10] | | Beta-lactam dosing in a critically ill patient | What does dose individualization mean? | A regimen informed by measured concentrations from that patient | [7] | | An isolate is carbapenem-resistant *A. baumannii* | For which organisms is cefiderocol preferentially supported? | NDM-producing Enterobacterales and *Stenotrophomonas maltophilia* | [3] | The *Culex pipiens* item is worth keeping in a vector biology unit, since reinforcement learning methods performed effectively with fewer features and the two bioclimatic variables dominated the model [8]. ## Choosing a Practice Question Format Format is a design decision, not a matter of taste. Each format measures something different. | Format | Best for | Limits | |--|--|--| | Single best answer, four options | Broad coverage, fast scoring, psychometrics | Guessing at 25% | | Extended matching | Differential diagnosis and organism lists | Slow to write, hard to blueprint | | True/false cluster | Rapid retrieval checks | High guessing, poor discrimination | | Short answer | Recall without cues | Scoring burden, rater variance | | Calculation item | MIC interpretation, dilution math | Only covers quantitative objectives | | Structured practical station | Bench skills, plating, Gram stain reading | Requires a trained observer | | Case vignette with sequential questions | Clinical reasoning across a scenario | Items are not independent | For a large microbiology course, single best answer plus a small number of calculation and short answer items covers the most ground per hour of writing time. The structured practical station deserves separate mention. Bench skills such as aseptic technique, colony picking, and Gram stain interpretation do not measure well on paper. A station with a scored checklist is the format that matches the objective. Sequential case questions have a statistical trap. If question 3 depends on getting question 1 right, the items are not independent, and the reliability estimate for the set is inflated. Use sequential cases for teaching sets and independent items for graded exams. ## Checking That the Questions Work: Item Analysis Item analysis is the routine of computing two numbers per item and reading the distractor counts. It requires no special software. ### Difficulty The difficulty index, usually written p, is the proportion of learners who answered correctly. ``` p = (number correct in upper group + number correct in lower group) / (total in both groups) ``` A p of 1.00 means everyone got it right and the item taught you nothing about the cohort. A p below 0.30 on a four-option item means the item may be miskeyed or the concept was not taught. ### Discrimination The discrimination index, written D, compares performance in the top and bottom groups, usually the upper and lower 27% by total score. ``` D = (correct in upper group - correct in lower group) / size of one group ``` D ranges from -1.00 to +1.00. Values of 0.30 and above are conventionally considered useful, values between 0.10 and 0.30 are marginal, and negative values mean high scorers got the item wrong more often than low scorers, which almost always signals a miskey or a trick. ### A worked example ``` n = 40 learners upper group (n = 12): 11 correct lower group (n = 12): 5 correct p = (11 + 5) / (12 + 12) = 16 / 24 = 0.67 D = (11 - 5) / 12 = 6 / 12 = 0.50 option counts, upper / lower A (key) 11 / 5 B 0 / 9 C 1 / 0 D 0 / 0 ``` Read the result in order. Difficulty 0.67 sits comfortably in the mid range. Discrimination 0.50 is strong. Option B attracted nine of twelve low performers and no high performers, which is exactly what a well-built distractor should do, since it is capturing a real misconception. Option D was selected by nobody in either group, so it is a dead option and should be replaced or the item converted to three options. Option C attracted a single high performer, which is noise at this sample size but worth watching across terms. Rewrite the dead option, keep the item, and recheck next term. A ten-item set with two dead options across the bank is normal on a first run. ### When the numbers disagree If p is high and D is low, the item is too easy to discriminate. Move it to a practice set. If p is low and D is high, the item is hard but valid. Keep it in an honors or graduate bank. If D is negative, stop and check the key before anything else. A miskey is far more common than a genuinely inverted item. ## Using the Set in a Microbiology Course A bank is a curriculum asset. How you deploy it changes what learners get from it. ### Spaced review Spaced repetition means revisiting material at expanding intervals rather than in one block. An introductory microbiology course that gave students instructor-made spaced repetition flashcards found overall exam performance unchanged (p = 0.2657) while students' confidence in their ability to succeed improved (p = 0.0066), along with their belief that the course made them think like a microbiologist (p = 0.0011) and that the approach was effective (p = 0.0076) [11]. The practical reading: spaced review changes how learners feel about the material faster than it changes their scores. Deploy it for retention and confidence, and do not assume a flashcard deck will lift exam averages on its own. ### Pre and post testing Use two parallel forms built from the same learning objectives. The clinical microbiology escape game study did exactly this, with two distinct 10-item tests assessing the same predefined objectives before and immediately after the activity [2]. Parallel forms require matched difficulty, so check p on both forms after the first administration and adjust. ### Micro-checks inside modules Short knowledge checks embedded in learning modules work well. In an introductory food microbiology course, 73% to 76% of students voluntarily accessed supplemental online modules with average engagement times of 5 to 8 minutes, and performance on module-related exam items improved by 6% to 10% relative to overall exam averages [12]. Short, replayable checks attached to a specific module are the format most likely to be used. ### Team sets Sets designed for teams of three to four, with a version adapted to the training level, let one activity serve undergraduates and postgraduates. The bioinformatics escape game ran on exactly that model, with 40% of participants being second-year medical students and 60% postgraduate MDs or PharmDs [2]. Team formats also expose reasoning, since a wrong answer has to be defended out loud. ### Simulation-based practice Where bench access is limited, simulated practice has measurable effects. A 14-week quasi-experimental study of 128 undergraduate food science students compared an AI-enhanced virtual laboratory with traditional laboratory instruction. The virtual laboratory group reached higher knowledge acquisition (86.3% versus 78.6%, p < 0.001, d = 0.80) and higher skill development (84.6% versus 72.8%) [13]. Simulated stations are a reasonable substitute for practice questions when the objective is procedural. ### Scaffolded self-directed preparation A curriculum-aligned custom GPT used for case preparation produced higher self-directed learning rubric scores (10.3 versus 9.4 out of 12, mean difference 0.9, 95% CI 0.5 to 1.3, p < 0.001, d = 0.71) and higher small group discussion scores (14.9 versus 13.7 out of 16, mean difference 1.2, 95% CI 0.7 to 1.7, p < 0.001, d = 0.78) among 44 second-year medical students, while delayed retention did not differ [14]. Scaffolding improves preparation quality. It does not automatically improve long-term retention, so pair scaffolded preparation with spaced review. ## Common Mistakes and Limitations **Answering the stem twice.** A stem that contains the diagnostic finding and a lead-in that asks for the diagnostic finding makes a recall item that looks like an application item. Fix by moving one finding into the options. **Unnamed distractors.** If you cannot write the misconception in a note before writing the option, the option is filler. Filler options are the main reason items have a p above 0.90. **Negative stems without emphasis.** "Which of the following is not a selective medium" produces errors that reflect reading speed, not microbiology knowledge. **Length cuing.** The key is often the longest and most qualified option because the writer wanted to be precise. Trim the key or pad the distractors to equal length. **Stale breakpoints.** Susceptibility items tied to a superseded breakpoint table become wrong silently. Store the breakpoint table version in the source field. **Treating AI-generated items as ready to use.** AI has improved microbial classification and detection and supports antimicrobial resistance prediction, but deep learning models often lack transparency, and overfitting and limited generalizability are recognized concerns, alongside data privacy and legal issues [15]. Every generated item needs a human key check against a primary reference and a written rationale before it enters a bank. **Using a prediction model as a result.** Models that predict antimicrobial susceptibility from routinely collected metadata have shown moderate test accuracy, in the range of 57% to 76% depending on the antibiotic [10]. That supports triage and surveillance, not a reportable result. **Ignoring the positive side of microbiology.** In a survey of 924 college students in the United States, students viewed viruses more negatively than bacteria and struggled to give examples of beneficial effects for both [9]. A bank that contains only pathogen identification reinforces that gap. Add items on industrial microbiology, normal flora, and beneficial applications. **Over-reading a single term's statistics.** Item statistics from 30 learners are noisy. A D of 0.20 may be 0.45 next term. Accumulate at least two administrations before retiring an item. **Confusing practice with assessment.** Practice sets should be harder and should include full rationales. Graded exams should be blueprinted and should not reuse practice items verbatim. **The limitation that applies to all of it.** No template replaces subject expertise or a current reference. Patient-specific decisions belong to the treating clinician and the local laboratory's validated method, and no question bank settles what those should be. ## Frequently Asked Questions ### How many practice questions should a microbiology course set include? Aim for 8 to 12 items per major content domain, which yields roughly 50 to 90 items for a full semester course. Build three times that many if the bank also serves practice, and retire at least 10% each term after reviewing item statistics. ### What is the best format for microbiology practice questions? Single best answer with four options is the workhorse for coverage and scoring, and context-based vignettes outperform isolated facts. Add calculation items for susceptibility interpretation and structured practical stations for bench skills. ### Should practice questions be grouped by organism or by concept? Group by concept and blueprint by domain. Organism-by-organism banks teach lists and hide the transferable principles, such as how a resistance mechanism changes a susceptibility result across several genera. ### How do I write a distractor that actually works? Name the misconception first, then write the option that a learner holding it would pick. If you cannot state the misconception in one sentence, the option is filler and should be replaced. ### Can I use AI to generate microbiology practice questions? Yes, as a drafting aid. Every generated item needs a human-written rationale for each option, a check of the key against a primary source, and a note on the misconception behind each distractor. Transparency and generalizability limits mean generated items cannot enter a bank unchecked [15]. ### What is a good difficulty index for a practice item? For a graded exam, a p between 0.50 and 0.80 with a discrimination index of at least 0.30 is a reasonable target. Practice items can run harder, since the rationale is the teaching component. ### How often should a learner revisit the same practice question? Space the revisits. Use short intervals during the unit and expanding intervals across the term, and change the surface details of the item each time so the learner recognizes the principle rather than the wording. Spaced flashcard use has been linked to improved confidence and perceived value in an introductory microbiology course [11]. ### Do practice questions improve exam scores? Not automatically. A spaced-repetition flashcard trial in an introductory microbiology course found exam scores unchanged while confidence and perceived value rose [11]. Practice questions improve outcomes when they are hard enough to expose misconceptions, carry full rationales, and are spaced rather than crammed. ## Sources 1. [Evaluating the quality of multiple-choice question pilot database: A global educator-created tool for concept-based pharmacology learning.](https://pubmed.ncbi.nlm.nih.gov/39268814/) 2. [BioDecryptors, a Bioinformatics Escape Game for Undergraduate and Postgraduate Learners in Clinical Microbiology: Prospective Before-and-After Study.](https://pubmed.ncbi.nlm.nih.gov/42987315/) 3. [Optimizing cefiderocol use in clinical practice.](https://pubmed.ncbi.nlm.nih.gov/42929061/) 4. [Benchmarking reveals superiority of deep learning variant callers on bacterial nanopore sequence data.](https://pubmed.ncbi.nlm.nih.gov/39388235/) 5. [Core Competencies of Nurses Working in Hospital Infection Prevention and Control: A Scoping Review and Cluster Analysis.](https://pubmed.ncbi.nlm.nih.gov/42319202/) 6. [Antimicrobial Peptide Databases and Prediction Tools: Toward a Standard Evaluation Framework.](https://pubmed.ncbi.nlm.nih.gov/43001936/) 7. [Consensus Guidance for Beta-Lactam Antibiotic Dose Individualization in Acutely Ill Patients: Endorsed by the American College of Clinical Pharmacy, European Society of Clinical Microbiology and Infectious Diseases, European Society of Clinical Microbiology and Infectious Diseases PK/PD of Anti-Infectives Study Group, Infectious Diseases Society of America, International Association of Therapeutic Drug Monitoring and Clinical Toxicology, Society of Critical Care Medicine, and the Society of Infectious Diseases Pharmacists.](https://pubmed.ncbi.nlm.nih.gov/42487243/) 8. [Comparing machine learning, deep learning, and reinforcement learning performance in Culex pipiens predictive modeling.](https://pubmed.ncbi.nlm.nih.gov/41231961/) 9. [Positive microbiology: addressing students' knowledge gaps regarding the benefits of microorganisms.](https://pubmed.ncbi.nlm.nih.gov/40742141/) 10. [Surveillance-Driven Machine Learning for Prediction of Antimicrobial Susceptibility: An Explainable Modeling Framework using the Pfizer ATLAS Dataset (2004 - 2023).](https://pubmed.ncbi.nlm.nih.gov/42823942/) 11. [Usage of Spaced Repetition Flashcards to Improve Educational Outcomes in Microbiology.](https://pubmed.ncbi.nlm.nih.gov/39507168/) 12. [The development of supplemental multimedia learning modules and their impact on student learning in food biotechnology courses.](https://pubmed.ncbi.nlm.nih.gov/39508351/) 13. [AI-enhanced virtual laboratories improve learning outcomes and student engagement in food microbiology education.](https://pubmed.ncbi.nlm.nih.gov/42156992/) 14. [Curriculum-Aligned Conversational AI to Support Pathology Self-Directed Learning in Preclinical Medical Students: Crossover Randomized Pilot Study.](https://pubmed.ncbi.nlm.nih.gov/42873617/) 15. [The benefits and biases of artificial intelligence in microbiology.](https://pubmed.ncbi.nlm.nih.gov/42826571/)