Accuracy and cut-off values of different acoustic measures in the sustained vowel task of Brazilian Portuguese speakers.

The design takes one snapshot in time, so it cannot say whether these scores change with treatment or whether the cut-offs hold up on a second assessment. Only 99 people without a voice disorder were compared with 277 who had one, and those 99 were university students and staff rather than people who came to a clinic. How well a cut-off performs depends on how common the condition is in the group tested, so these numbers may behave differently in your caseload. There is no agreed gold standard for what counts as a voice disorder. The researchers used a doctor's laryngeal examination combined with three speech-language pathologists' listening judgements, and a listening judgement is subjective even though repeat ratings agreed well. Professional voice users and anyone who had already had voice therapy or surgery were excluded, so the cut-offs are untested in exactly the patients you might most want them for. Forty-four measures were extracted but only the 23 that reached the researchers' own threshold are reported, so a reader cannot see how the other 21 performed or whether the cut-offs were chosen after looking at the data. The paper recommends SNL 100-2600Hz, SNL 100-3000Hz and SNL 100-8000Hz for suspected cases on the grounds that both sensitivity and specificity reach 80%, but its own table gives SNL 100-2600Hz a specificity of 72.12% and SNL 100-8000Hz 70.63%.

The authors declare no financial support and nothing to declare. No funder or company is named as having paid for, designed or written up the work.

Do not treat a positive result on any of these measures as a diagnosis of a voice disorder. The paper's own numbers show a positive result was correct only 17.5% to 31.6% of the time, while a negative result was reliable at 94.1% to 100%. The cut-offs also come from one clinic where just 99 of 376 people had healthy voices, most of them university students and staff, so they may not transfer unchanged to your caseload.