This is a subtle question and pervasive concern. Also, it is challenging to provide a concrete response without a sense of the particular application you had in mind.

Trivially, anyone you interview or measure has to some degree "survived" long enough to participate in your study or to obtain the specifics of their circumstances that interest you. Scientific American had a fun but relevant comment on how this manifests in everyday "miracles" (not health science related): https://www.scientificamerican.com/article/how-the-survivor-bias-distorts-reality/

The point is actually profound. This bias cannot be "avoided", it can only be contemplated and responded to when appropriate. One can also regard it as a kind of censoring for which methods are well documented.

The more proximal question is whether "survivors" (or participants) differ from "non-survivors" (non-participants) in ways that are relevant to your topic of study or to the specific question you are asking. If so, then one can investigate whether there any additional data you might be able to collect to either control for potential bias or to triangulate against an outside data source (e.g., hospital records, census, mortality reports, etc.) for an assessment of impact.

A classic manifestation that offers some relevant insights - by Cohen and Cohen - refers to the "clinician's illusion" and relates to the potential for bias between an incidence sample and a prevalence sample. It relates to survivor bias in that prevalent cases may not survive to be in your study, whereas one must be alive to become an incident case. See: https://www.ncbi.nlm.nih.gov/pubmed/6334503

Google "clinicians illusion" and follow the myFSU link to get a PDF if you don't have institutional access to the full publication. Also "Survivor bias and risk assessment" http://dx.doi.org/10.1183/09031936.00094112

Your question as to ways to avoid this in data collection depends mainly on the specific situation at hand. Basically, it is almost impossible to avoid but generally fairly easy to anticipate. A meta-question might be: are there people who are less likely (or definitely wouldn't) participate in my study who might hold a key part of the answer I am seeking and who, if I don't account for them, might lead me to make an incorrect conclusion?

If the answer is yes then one might be able to characterise who they are vis a vis who actually is in your sample and assess the potential impact of their exclusion (e.g., collecting basic demographics on refusals, assessing effect of exclusion criteria, recording date of onset or time with condition, etc.). One might also use an a priori causal model of the phenomenon you're studying to identify (or hypothesize) factors that could impact survival or non-survival and collect data on these factors for use in analysis to either control out their effects or to explore their potential causal role (depending on the specifics of your analysis methods and research questions).

At worst, one might simply account for survivor bias by constraining one's claims to the narrowest set of conditions that are true for the sample one has, but to do so can be overly conservative and perhaps uninformative. With some careful thought and thoughtful planning it is almost always possible to do better.

When I looked at your foundation's website, it seems like the work you do is the kind of work where learning about who is not able to take advantage of your programs and what their barriers are to access could be highly informative.

I hope this response is useful at some level. Perhaps ask again if you do have a specific research question and concrete application as then a more technical response could be forthcoming.

Good literature should have been found by you, by now.

May read:1. STANDARD DEVIATIONS; by Gary Smith. Duckworth Overlook (25. September 2014). Kindle Edition;

2. 'Surviving the Survivor Bias' [  https://www.valueresearchonline.com/story/h2_storyView.asp?str=27798  ]