The review of literature is often written last and should be drafted first. Its job is not to show that you have read widely. Its job is to make one gap so obvious that your research question reads as the only reasonable next sentence.
What the chapter is actually for
An examiner reading your review is asking one question: given everything already published, does this study need to exist? Everything in the chapter either answers that or is padding. A summary of forty studies answers nothing. A chapter showing that three groups have measured something in Western populations, two have measured it in India, none in the population you see in your outpatient department, and that the two Indian studies disagree, answers it in a paragraph.
So write towards a gap you can state in one sentence before you write anything else. If you cannot state it, you are not ready to write the chapter; you are still reading. Say it out loud to your guide first. 'Nobody has looked at this in patients already on the standard drug' is a gap. 'More research is needed' is not.
Searching PubMed properly
The usual approach is to type a phrase into PubMed, take the first page of results and stop. PubMed accepts that, because it maps your free text onto its own indexing vocabulary automatically, but in doing so it has decided on your behalf what you meant and you never see what it discarded. The National Library of Medicine documents this as automatic term mapping. Getting control back means building the search yourself.
Medical Subject Headings (MeSH) are the terms NLM indexers attach to each article. Look your concept up in the MeSH Database rather than guessing it, because the indexed heading is often not the word clinicians use at the bedside. A MeSH heading explodes by default when searched on PubMed, which means it also retrieves the narrower headings beneath it in the tree. For a thesis review that is usually what you want, occasionally far too much.
Then build in concepts rather than sentences. One line per concept, synonyms joined by OR; concepts joined by AND; free-text spellings included alongside the MeSH heading, because an article indexed last week may not carry the heading yet. Field tags restrict where PubMed looks, [mh] for the MeSH heading and [tiab] for title and abstract. Note one trap: forcing a phrase with quotation marks, or truncating a word, switches automatic mapping off for that term, so quoting everything quietly narrows your search instead of tightening it.
Keep the final search string. Paste it into your methods with the database, the date you ran it and the number of records it returned. It costs two lines and it is the cheapest way to show that your search was systematic rather than anecdotal.
- Build one line per concept, then combine the lines with AND
- Look every concept up in the MeSH Database instead of guessing the heading
- Add free-text forms with [tiab] so recent, not-yet-indexed articles are not lost
- Run the same concepts in Embase or Scopus if your institution subscribes, and check the Indian journals in your field directly, because PubMed does not index every one of them
- Record the search string, the database, the date and the hit count
Medical literature review: a search strategy
The paragraphs above explain the principles. This section is the working sequence, one step per line, each with what to do and a worked example built on a single thesis question: does early enteral feeding shorten hospital stay after elective colorectal surgery in adults. Substitute your own concepts and keep the shape. The PubMed syntax is from the National Library of Medicine's PubMed User Guide; the Europe PMC syntax was checked against Europe PMC's own search service.
- Step 1, frame the question: write the question as population, intervention, comparison and outcome before touching a database, and keep only the concepts the search needs, usually two or three: example, adults after elective colorectal surgery (population), early enteral feeding (intervention), length of hospital stay (outcome); the comparison, delayed feeding, is left out of the search because the retrieved papers define it themselves
- Step 2, find the MeSH heading for each concept: open the MeSH Database, search the lay term, read the scope note and the tree position, and note the heading and whether it has narrower terms beneath it; the PubMed User Guide states that a MeSH term automatically includes the more specific terms below it, which is usually what a thesis review wants: example, Colorectal Surgery as a heading and Enteral Nutrition as a heading, with Length of Stay for the outcome
- Step 3, add free-text synonyms for each concept: list the words authors actually use in titles and abstracts, including British and American spellings and abbreviations, because an article indexed last month may not yet carry its heading, and tag each with [tiab]: example, colorectal surgery, colectomy, colon resection, rectal resection, bowel surgery
- Step 4, build one line per concept with OR: join the MeSH heading and its free-text forms into a single bracketed line, so each line retrieves everything about that one concept: example, ("Enteral Nutrition"[mh] OR "enteral feeding"[tiab] OR "early feeding"[tiab] OR "early oral feeding"[tiab])
- Step 5, combine the concept lines with AND, in uppercase: the PubMed User Guide requires Boolean operators in uppercase and processes the search left to right, nesting with parentheses where a unit must be processed first; NOT narrows aggressively and removes any article that mentions the excluded term even in passing, so use it rarely: example, (line 1) AND (line 2) AND (line 3)
- Step 6, use truncation and phrases deliberately: an asterisk substitutes for zero or more characters at the end of a term, and double quotes force a phrase; both bypass automatic term mapping for that term, so use them to tighten a line you have already built with MeSH, not as a shortcut to building it: example, colectom*[tiab] retrieves colectomy and colectomies
- Step 7, set the date limit last and record why: a date range is entered with a colon between the two dates followed by the publication date tag, and the PubMed User Guide notes that [dp] covers both print and electronic dates; choose the range for a reason you can write in the methods, such as a change in surgical practice or a landmark trial, not to shrink the number: example, AND 2010:2025[dp]
- Step 8, run the same concepts in Europe PMC: it is a free second index that covers PubMed abstracts and PubMed Central full text and also indexes preprints, which PubMed does not; its syntax is field name, colon, term, with ranges in square brackets: example, (TITLE:"enteral feeding" OR ABSTRACT:"enteral feeding") AND PUB_YEAR:[2010 TO 2025], and MESH:"Enteral Nutrition" for the indexed heading; SRC:PPR restricts to preprints, which belong in your review only as unreviewed work and labelled as such
- Step 9, check what the search missed: read the reference lists of the five most relevant papers and the papers that cite them, search the Indian journals in your field directly if they are not indexed, and add any missing term you discover back into the relevant concept line and re-run, so the final string is the one that found everything you cite
- Step 10, record the search as you ran it: paste the exact final string, the database, the date it was run and the number of records returned into a file on the same day, and then into the methods chapter; the PubMed User Guide notes that the Advanced search history is limited to the last 100 searches and expires after eight hours of inactivity, so the record has to be yours, not the website's: example, PubMed, the date run, the string as above, and the record count exactly as the screen showed it, before any screening
- Step 11, screen and keep the count at each stage: export every record to a reference manager, remove duplicates across databases, screen titles and abstracts against the inclusion criteria you wrote in Step 1, then full texts, and note how many were removed at each stage and why; the counts are what a PRISMA-style flow of the search looks like if your guide asks for one, and they cost nothing if kept as you go
Two cautions. A search that returns forty records is not necessarily better than one that returns four hundred; it may simply have missed a synonym. And an identical string run on two different days returns different counts, because the databases grow, so the date is part of the record and not decoration.
Reading for the gap, not for the findings
There is a difference between reading a paper and mining it. For a review of literature you are mining. For every study that survives your screen, record one row: design, setting and country, sample size, who was included, what was measured and how, the main result, and the column that does the real work, what it could not answer. That last column is your gap, assembled one study at a time.
Do this in a spreadsheet from the first paper, not in your head and not in a folder of PDFs. Twenty rows of that table is a review of literature. Ninety unread PDFs is a weekend you will lose twice.
Read the methods before the discussion. A discussion tells you what the authors wish they had shown. The methods tell you what they did, and the limitations they admit are usually the exact space your study occupies.
Structure: general, then specific, then you
Order the chapter as a funnel. Open with the condition and its burden, what it is, who it affects, why it matters clinically. Narrow to the specific aspect you are studying: the intervention, the marker, the technique, the scoring system. Then the evidence on that aspect. Then Indian and regional data specifically, and whether it agrees with the rest. Close with a short paragraph stating what remains unanswered and therefore what this study will do.
Group by theme, never by author or by year. A chapter that runs 'one author found, then another found, then a third found' is a bibliography with paragraph breaks. A chapter that runs 'three hospital-based studies report higher values in this group, though all three recruited from tertiary centres' is a review. Studies are evidence for your argument, not the subject of it.
The last paragraph of the review and the first paragraph of your aims should read as consecutive sentences. If they do not, one of the two is wrong.
Cite while you write, never afterwards
Citing at the end is how reference lists go wrong: numbers drift out of order, two papers merge into one, page ranges get typed from memory. Insert the citation in the same keystroke as the claim. Use a reference manager and let it renumber: Zotero is free and open source, and Mendeley has a free account tier. Vancouver style[4] numbers references in order of first appearance, which is exactly the order a manager maintains and exactly the order manual editing destroys.
Pull each reference from the source record rather than from another paper's reference list, because errors in reference lists propagate. And read what you cite. An examiner who opens one reference and finds it says something other than your sentence claims will open five more.
How many references is reasonable
There is no national rule. Neither the National Medical Commission nor the National Board of Examinations in Medical Sciences[2] publishes a number, so any figure you are given is a local convention. Ask your department what its recently accepted theses looked like; that answer is worth more than any number here. Two real constraints do exist. The NBEMS thesis guidelines cap a DNB submission at 80 pages, which bounds the list whether you like it or not, and every reference has to be one you have actually read.
The honest test is coverage rather than count. If a reader who knows the field could name a landmark paper you have not cited, you are short. If you have cited forty papers and could not say what six of them concluded, you are long.
What makes an examiner mark it down
- A chapter that summarises studies one by one and never states a gap
- Textbook-level background with no primary literature behind it
- No Indian or regional data, in a thesis conducted in an Indian hospital
- Claims with no citation, or a citation that does not support the claim made
- References out of Vancouver order, or listed in the bibliography but never cited in the text
- Studies cited from secondary sources, with the original never opened
- Silence on the one contradictory study the examiner happens to know
- Heavy overlap with published prose, because the chapter was assembled from sources instead of written from an evidence table
That last one solves itself if the chapter comes out of your own evidence table, because you are then describing your reading rather than reproducing it.