Most MeSH terms are assigned multiple allowable subheadings.  Citations which are assigned these MeSH headings are also assigned appropriate subheadings, based upon the content of the article.

For example, the MeSH term Angina Pectoris is assigned as allowable  subheadings /di (diagnosis), / th (therapy), /ep (epidemiology), and so on.   However  /tu (therapeutic use), is not allowed.      Any citation assigned the MeSH term Angina Pectoris could be assigned any or multiple allowable MeSH subheadings, as indicated by the content of the article.

Subheadings help limit your search to citations for your particular interest.  Thus they are another mechanism for limiting (filtering) your search.   In general we do not recommend using subheadings to limit a search.   Remember, subheadings (and MeSH terms) are assigned by human readers.  Subheadings are notoriously unreliable (Greenhalgh).  You are likely to miss important citations if you overdo it with subheadings.  Furthermore, subheadings, unlike a search  filter, are not designed to filter for study quality.  Search filters, depending on the specific filter,  screen for citations which are randomized, blinded, or employ specific designs or outcome measures.

To see an example, consider the following:

You are evaluating a patient who presents with syncope.  The patient happens to be taking terfenadine, a "non-sedating" antihistamine.  Turning to OVID, you seek evidence on the risk of syncope in patients taking antihistamines.   You perform the following search using the HealthStar database.

Line 1: you performed a search of syncope limited to the subheading etiology (ET)
Line 2: after examining the tree for terfenadine, you built up a search related to this group of antihistamines
Line 3: Boolean AND combination of your disease (syncope) and your exposure (antihistamines)
               This yields 6 citations, none were relevant

Line 4: backtracking, you run a search filter for etiology or cause 
Line 5: now search the disease as a key word (mp)
Line 6: Boolean AND combination of the search filter, disease, and the exposure
                This yields 17 citations.  Two of these are pertinent including a large HMO based cohort study

The problem with the first search was choosing the wrong subheading for syncope, an easy mistake.  In the cohort study that we found, syncope carried the subheadings CI (chemically induced) and EP (epidemiology)

Similarly, had you searched antihistamines/co (complications) you would  not have found anything.  This is not an allowable subheading for antihistamines.  To find the allowable subheadings use the "tools" function in ovid.  The subheading used in this article for antihistamines was AE (adverse effects).

To become familiar with the indexing take a look at the "complete reference" for articles which are relevant to your search.  You can see exactly how the article is referenced.

Below is the indexing from the complete reference for the citation which answered our question best

Terfenadine-associated ventricular arrhythmias and QTc
interval prolongation. A retrospective cohort comparison
with other antihistamines among members of a health
maintenance organization.

Annals of Epidemiology. 5(3):201-9, 1995 May.