Context. Broad absorption lines (BALs) in the spectra of quasi-stellar objects (QSOs) originate from outflowing winds along our line of sight; winds are thought to originate from the inner regions of the QSO accretion disk, close to the central supermassive black hole (SMBH). These winds likely play a role in galaxy evolution and are responsible for aiding the accretion mechanism onto the SMBH. Several works have shown that BAL equivalent widths can change on typical timescales from months to years; such variability is generally attributed to changes in the covering factor (due to rotation and/or changes in the wind structure) and/or in the ionization level of the gas. 

Aims. We investigate BAL variability, focusing on BAL disappearance. 

Methods. We analyze multi-epoch spectra of more than 1500 QSOs - the largest sample ever used for such a study - observed by different programs from the Sloan Digital Sky Survey-I/II/III (SDSS-I/II/III), and search for disappearing C IV BALs. The spectra cover a rest-frame time baseline ranging from 0.28 to 4.9 yr; the source redshifts range from 1.68 to 4.27. 

Results. We detect 73 disappearing BALs in the spectra of 67 unique sources. This corresponds to 3.9% of BALs disappearing within 4.9 yr (rest frame), and 5.1% of our BAL QSOs exhibit at least one disappearing BAL within 4.9 yr (rest frame). We estimate the average lifetime of a BAL along our line of sight (approximate to 80-100 yr), which appears consistent with the accretion disk orbital time at distances where winds are thought to originate. We inspect various properties of the disappearing BAL sample and compare them to the corresponding properties of our main sample. We also investigate the existence of a correlation in the variability of multiple troughs in the same spectrum, and find it persistent at large velocity off sets between BAL pairs, suggesting that a mechanism extending on a global scale is necessary to explain the phenomenon. We select a more reliable sample of disappearing BALs on the basis of some criteria adopted in a previous publication, where a subset of our current sample was analyzed, and compare the findings from the two works, obtaining generally consistent results.